Behaviorism vs Cognitivism vs Constructivism: Which to Use

Go from storyboard to an interactive course in minutes

Start Free

Picture an energy retailer's contact centre (a hypothetical example) training a new intake of agents to handle refund requests on the phone. One designer wants drills on the CRM screens. Another wants the refund policy explained properly, with examples. A team leader wants the new agents working through the messy cases that reach her desk every week. Each of them is right about a different part of the call, and a course built on only one of their ideas leaves the other parts untrained.

I'm an instructional designer at Coursebox, where we make instructional design software that turns a designer's brief into a course outline, so I have a stake in how courses get planned. Those three views come from three learning theories: behaviorism, cognitivism and constructivism. I use the refund course to show all three.

What are behaviorism, cognitivism and constructivism?

Behaviorism, cognitivism and constructivism are the three classic learning theories in instructional design: behaviorism treats learning as a change in observable behaviour, cognitivism as a change in organised knowledge, and constructivism as meaning that learners build for themselves from experience.

View the full-size diagram

They are theories of learning, meaning explanations of how people learn, and instructional designers use them to choose teaching strategies. The canonical comparison for designers is Peggy Ertmer and Timothy Newby's article, first published in Performance Improvement Quarterly in 1993 and reprinted in 2013 with an update. Page numbers here are from the 2013 reprint. Their one-line definitions are:

• Behaviorism "equates learning with changes in either the form or frequency of observable performance" (p. 48).
• Cognitivism treats learning as "discrete changes between states of knowledge rather than with changes in the probability of response" (p. 51).
• Constructivism "equates learning with creating meaning from experience" (p. 55).

The three arrived in that order. Behaviorism was dominant when instructional design began around 1950, so in Ertmer and Newby's words "ID has its roots in behavioral theory" (p. 48). The field shifted toward cognitive science in the late 1950s, and the authors added constructivism as a third category "because of its recent emphasis in the instructional design literature" (p. 46). All three share a starting definition of learning. The authors argue that the differences "lie more in interpretation than they do in definition" (pp. 45 to 46), and that the theories "will often prescribe the same instructional methods for the same situations" under different names (p. 46).

Behaviourism, the British and Australian spelling, is the same theory. I use the US spelling for the three theory names and Australian spelling everywhere else.

Who developed each theory?

No single person founded any of the three. Each grew from several researchers' work, and the instructional design versions were assembled later.

A few names get mixed up often:

• Thorndike, Pavlov and Skinner did different things. The law of effect is Thorndike's (1911, p. 244). Classical conditioning, where a response becomes tied to a new signal, is Ivan Pavlov's paradigm. Skinner worked on reinforcing the learner's own responses. In a 1958 article on teaching machines he wrote that by arranging "contingencies of reinforcement" specific behaviour can be "set up and brought under the control of specific classes of stimuli" (p. 970). Watson's 1913 manifesto set the programme of studying behaviour alone.
• Piaget belongs with constructivism in instructional design. Ertmer and Newby name constructivism's roots as "Piaget, Bruner, and Goodman" (p. 55, citing Perkins). Piaget set his idea of learning directly against behaviorism: "the fundamental relation is one of assimilation. Assimilation is not the same as association" (Piaget, 1964, p. 185). He is the founding figure of cognitive constructivism, which is a different thing from the information-processing cognitivism of Miller, Atkinson and Shiffrin.
• Vygotsky is the main figure of social constructivism. In Mind in Society he wrote that every function in development "appears twice: first, on the social level, and later, on the individual level" (Vygotsky, 1978, p. 57). He defined the zone of proximal development as the distance between what someone can do alone and what they can do "under adult guidance or in collaboration with more capable peers" (p. 86). The phrase "more knowledgeable other", often attached to it, does not appear in the book.

Later theories extend the list. Connectivism addresses learning across networks of people and tools. Albert Bandura's social cognitive theory explains observational learning, self-efficacy, agency and the reciprocal influence of personal factors, behaviour and environment.

Ertmer and Newby's comparison: what it says and which version to read

Ertmer and Newby compare the theories by asking each one the same seven questions. Five come from Dale Schunk's 1991 textbook, and the authors added two for designers (p. 46):

1. How does learning occur?
2. Which factors influence learning?
3. What is the role of memory?
4. How does transfer occur?
5. What types of learning are best explained by the theory?
6. What basic assumptions or principles of this theory are relevant to instructional design?
7. How should instruction be structured to facilitate learning?

The definition of learning they start from is Thomas Shuell's, as interpreted by Schunk: "an enduring change in behavior, or in the capacity to behave in a given fashion, which results from practice or other forms of experience" (p. 45). The authors do not pick a winner. They write that they have "consciously chosen not to advocate one theory over the others", and that designers should choose methods "on the basis of information gathered about the learners' present level of competence and the type of learning task" (p. 61). Mixing theories this way is what they call systematic eclecticism, borrowing Glenn Snelbecker's term, with a warning that "to be an eclectic, one must know a lot, not a little, about the theories being combined" (p. 62).

The 2013 update kept that position. The authors wrote that "the answers to the seven organizing questions remain largely the same" (p. 65), and that the designer's job is still to understand "the strengths and weaknesses of each learning theory" and select strategies to match (p. 69). What changed was the setting, such as the internet, informal learning and the spread of constructivist methods like problem-based learning.

• Read the 2013 reprint in Performance Improvement Quarterly 26(2), pages 43 to 71. It contains the full 1993 article and the five-page update, so it is the current version.
• Use the free 1993 text on EdTech Books, an openly licensed reprint, if you cannot get the journal. The page numbers differ from the 2013 reprint.
• Then read both sides of the guidance debate: Kirschner, Sweller and Clark (2006) and the reply from Hmelo-Silver, Duncan and Chinn (2007).

One refund course designed three ways

The worked example is a phone refund course at an energy retailer's contact centre. The learners are new agents in their first weeks, some with contact-centre experience elsewhere, none with this retailer's CRM or refund policy. On a refund call, an agent has to do three things: log the request in the CRM with the fields in the required order, decide whether the request meets the policy for a standard refund, and work out what to do when a case does not fit the policy neatly, such as a customer billed twice after moving house. Agents get the refund policy, the CRM screens and the customer's account history to work from, and the intended outcome is that standard refunds go through correctly and unusual ones reach a sound, explained decision.

Designing the same refund call three ways shows what each theory sees, and what it misses.

1. Behaviorism: drill the routine

A behaviorist designer treats the refund call as a set of correct responses to practise until they are automatic. Instruction is "structured around the presentation of the target stimulus and the provision of opportunities for the learner to practice making the proper response", with cues to prompt the response and reinforcement to strengthen it (Ertmer and Newby, p. 50).

• Sample goal: Enter every refund request in the CRM with the fields in the required order, with no errors across three practice rounds in a row.
• How to assess it: Practice rounds marked right or wrong, against a mastery standard set before the course starts.
• In the course: Field prompts on screen at first, then repeated rounds with immediate right or wrong feedback, with the prompts removed as accuracy rises.

Drill supports the CRM routine because the required field sequence is fixed in this example. That practice alone gives agents little preparation for a refund the policy does not cover. These strategies are useful emphases, not methods owned exclusively by one theory.

Case study: Safety skills training for new hospital staff in Toronto

The Centre for Addiction and Mental Health (CAMH) in Toronto, which the researchers describe as Canada's largest mental health hospital, teaches every new clinical hire self-protection and team-control skills during a mandatory two-week orientation, for use as a last resort when a patient becomes violent. Its usual training followed a describe, demonstrate and do pattern.

• What they did: The researchers tested behavioural skills training, a method from applied behaviour analysis that loops instruction, modelling, practice and feedback until each person reaches a set standard of 80% competence, against the usual training in a pragmatic randomised trial.
• Evidence and result: In the 2024 trial report, of 360 orientation attendees, 199 consented and 99 completed all three assessments (55 in the new training, 44 in the usual one). Observers scoring video found the new group's skills higher straight after training (Cohen's d = 1.45 for self-protection and 2.55 for team control) and still higher at one month (d = 0.82 and 0.62), while confidence rose equally in both groups.
• Limitation: Skills were scored in a training assessment, nobody measured behaviour on the ward, and half the participants missed the one-month follow-up. By one month, the new group's competence scores had fallen to 66% and 73% of their post-training level.
• Lesson: Practice with feedback to a set standard built more skill than describe, demonstrate and do, but skills in both groups faded within a month. The authors conclude that a single session without on-the-job feedback or booster sessions "may not be sufficient", so schedule refresher practice for any drilled routine.

2. Cognitivism: explain the rules

A cognitivist designer treats the refund call as information the agent must organise well enough to apply. In Ertmer and Newby's summary, "Instruction must be based on a student's existing mental structures, or schema, to be effective" (pp. 53 to 54), using strategies such as outlines, summaries and advance organisers (p. 53). Practical cognitive learning theory examples show the same methods in other settings.

• Sample goal: Decide whether a request meets the policy for a standard refund, and name the policy rule behind the decision.
• How to assess it: New standard cases the agent has not seen, where the agent decides and names the rule, marked against the policy.
• In the course: The refund policy explained in short chunks, a worked example of one decision from start to finish, and a one-page summary of the rules for agents to keep.

Explanation handles the policy well, because the rules are defined and the cases vary. It gives agents less to go on when two rules point different ways.

Case study: Apprentice training at a Sydney company

First-year trade apprentices at a Sydney company learnt soldering and industrial wiring from technical diagrams with written explanations. Diagrams with separate explanations force learners to split their attention between the two, which researchers from the University of New South Wales argued overloads working memory. They used cognitive load theory, a cognitivist account of those limits, to test how the explanations should be presented.

• What they did: In about an hour of computer-based instruction from the company's soldering module, apprentices saw a diagram with on-screen text, a diagram with the same text narrated, or a diagram with both at once.
• Evidence and result: In the 1999 study by Kalyuga, Chandler and Sweller, 34 apprentices were split into groups of 12, 11 and 11. On fault-finding and multiple-choice tests, "In all cases, means favoured the Audio text group", and narration plus the same text on screen was "the worst of the three instructional procedures".
• Limitation: The groups were small, the lesson lasted about an hour, the tests came straight afterwards, and nothing was measured later or on the job. The company is not named, and the study dates from 1999.
• Lesson: How an explanation is presented changes how much of it people learn. When a policy has a diagram or a process map, narrate it or show the text beside it, and avoid reading on-screen text aloud word for word.

3. Constructivism: work through the messy case

A constructivist designer treats the refund call as a problem agents make sense of through realistic cases and discussion. Ertmer and Newby describe the goal as having learners "elaborate on and interpret information" (p. 56), through strategies such as "situating tasks in real world contexts", "social negotiation" and "reflective awareness" (p. 58). The types and principles of constructivism, and how they apply to e-learning, are a subject of their own in constructivism learning theory.

• Sample goal: Resolve a refund case the policy does not settle cleanly, such as a customer billed twice after moving house, and justify the decision to the customer and a team leader.
• How to assess it: A new unusual case with its account history, where the agent writes a decision and reasoning, judged by a team leader against set criteria: the account history checked, the policy rule cited, and a clear next step for the customer.
• In the course: The double-billing case with the account history, both bills and the move dates supplied, then comparing the decision with a colleague's and reflecting on where the two differ.

An open case builds judgement, but only for agents who already know the policy and the CRM well enough to reason with them. Given to a new agent on day one, the same case is a guessing game.

Case study: A police academy redesign in Washington State

In 2009, Washington State's Basic Law Enforcement Academy moved from a lecture-based, quasi-military approach to problem-based learning, a constructivist method. Its aim was "to better teach critical thinking and field-based problem-solving skills, while still conveying the required legal, administrative, policy, and protocol knowledge."

• What they did: The academy rebuilt its academic curriculum around realistic problems, and kept drill where drill fitted. The evaluation names topics that "should still utilize the Traditional behaviorist techniques", such as firearms, defensive tactics, driving and physical training.
• Evidence and result: A 2010 Washington State University thesis compared the last two lecture-based classes with the first two problem-based classes, 115 graduates in all. The problem-based classes averaged 89.86 across academy tests against 87.95, and 88.05 against 84.61 on scored mock scenes. Satisfaction went the other way. Of the 53 graduates who answered a survey, 16 of 21 lecture-based graduates (76.2%) and 9 of 32 problem-based graduates (28.1%) agreed the curriculum was "well tailored to their immediate future in patrol".
• Limitation: This was a non-randomised comparison of two classes per approach in a master's thesis, with different instructors and cohorts. The scores come from academy tests, and performance on patrol was not measured.
• Lesson: A constructivist redesign can lift scores on realistic scenarios while trainees like it less. Keep drill for fixed routines, and expect trainees to need convincing that the open problems are worth the effort.

Side by side, each version fits one part of the call best: drill for logging the request, explanation for the policy, and the open case for the exceptions. Combined in that order, they make the course I would build.

Which theory should you use?

Use the theory that fits the learner and the task, and expect most courses to need more than one. Ertmer and Newby put the designer's question this way: "the critical question instructional designers must ask is not 'Which is the best theory?' but 'Which theory is the most effective in fostering mastery of specific tasks by specific learners?'" (p. 61). They answer it on two dimensions.

• How much the learner already knows about the task. Drawing on Donald Schön, they describe people moving from knowing the standard rules and facts of a profession ("knowing what"), to applying them to problem cases ("knowing how"), to "reflection-in-action" when familiar categories fail (p. 60).
• How much thinking the task demands. Tasks such as "basic paired associations, discriminations, rote memorization" suit behaviorist strategies, "classifications, rule or procedural executions" suit cognitive ones, and "heuristic problem solving" suits constructivist ones (pp. 60 to 61).

Their rule combines the two: "a behavioral approach can effectively facilitate mastery of the content of a profession (knowing what); cognitive strategies are useful in teaching problem-solving tactics where defined facts and rules are applied in unfamiliar situations (knowing how); and constructivist strategies are especially suited to dealing with ill-defined problems through reflection-in-action" (p. 60). They also report David Jonassen's view that introductory knowledge is better served by behaviorist and cognitive approaches, with a move to constructivist ones as learners gain the knowledge to handle complex problems (p. 57).

Cognitive load research later offered a mechanism, which Kalyuga, Ayres, Chandler and Sweller called the expertise reversal effect: "Instructional techniques that are highly effective with inexperienced learners can lose their effectiveness and even have negative consequences when used with more experienced learners" (2003, p. 23). A new agent has no stored knowledge of the policy to steer them through an open case, so step-by-step guidance and worked examples carry the load. An experienced agent already holds that knowledge, and the same detail becomes redundant work. The authors' conclusion is that "instructional design should be tailored to the level of experience of intended learners" (p. 29), and their review recommends that worked examples be "gradually faded out" as expertise grows (p. 27). For the new agents, that means drill and worked examples first, and the open case once they know the policy.

How to plan a course that uses all three

Plan the approach objective by objective, from what learners already know and what the task demands. This is the order I work in:

1. List the objectives from the job, as things people do, such as "log a refund request".
2. For each objective, find out how much learners already know about that task. Ask, test or look at their work, and do not assume it from a job title.
3. Decide what kind of task each objective is: a fixed routine, a rule applied to new cases, or a problem with no single answer.
4. Choose the approach to lead with, from the learner's prior knowledge and the type of task.
5. Write the assessment to match: right-or-wrong items with a mastery standard for a routine, new cases marked against the rule for a cognitivist objective, and a new open case judged against criteria for a constructivist one.
6. Plan the feedback for each approach.
7. Plan how guidance fades as learners gain experience, and schedule refresher practice for drilled routines.

Feedback differs by theory as well. In Ertmer and Newby's summary, "A behaviorist uses feedback (reinforcement) to modify behavior in the desired direction, while cognitivists make use of feedback (knowledge of results) to guide and support accurate mental connections" (p. 53). A right or wrong mark suits the CRM drill, while a policy decision needs feedback that explains which rule applies and why. If your course also needs a model for writing the objectives, Bloom's taxonomy classifies the thinking each objective asks for.

The refund course, planned objective by objective

Which approaches do course requests ask for?

To see how people brief courses, we classified a sample of requests typed into Coursebox in the 90 days to 7 October 2026: one per person, from English-language browsers, with no uploaded file, and with pre-outline answers and template duplicates removed. A language model classified the 266 prompts that remained. 101 were course requests, and we spot-checked the results by hand. The classifier counted generously, so these are model-coded estimates, not audited upper bounds or estimates of all Coursebox users.

Explanation and structure, the cognitivist approach, came up in about 63 of the 101 requests. Exploration, real problems and reflection came up in about 38, and drill, repetition and right-or-wrong feedback in about 34. Very few asked for the three together.

If your course has routines, rules and messy cases, as the refund course does, write the approach for each objective into the brief.

A few support threads point the same way. In a sweep of support threads in October 2026, 4 course creators told us in their own words that AI-generated drafts came out too text-heavy when they had asked for practice and exercises. A few drill requests also ran into limits: support confirmed in writing that Coursebox is not a spaced-repetition engine and has no drill block that adapts difficulty, so plan spaced review for drilled routines in another tool, such as your LMS.

Criticisms and limits of each theory

Each theory has a well-sourced weak spot, and the sharpest criticism is aimed at how constructivism gets turned into teaching.

• Behaviorism stops at observable responses. Ertmer and Newby note that it is "generally agreed that behavioral principles cannot adequately explain the acquisition of higher level skills", such as problem solving and critical thinking (p. 49). Noam Chomsky's 1959 review of Skinner's book Verbal Behavior made the case for language, calling the gaps "an indication of how little is really known about this remarkably complex phenomenon".
• Working-memory estimates are not a rule for the number of policy points on a screen. Miller described capacity in chunks, and later research has reported smaller estimates under different conditions. A chunk's size depends on what learners already know. Group the refund policy into meaningful decisions, remove irrelevant detail and test whether agents can use the rules, rather than impose a universal four-item limit.
• Constructivism describes learning better than it prescribes teaching. Kirschner, Sweller and Clark (2006) argue that "The constructivist description of learning is accurate, but the instructional consequences suggested by constructivists do not necessarily follow" (p. 78). Richard Mayer calls the leap from active learning to pure discovery the "constructivist teaching fallacy" (2004, p. 15).
• Guided constructivist methods hold up better. Hmelo-Silver, Duncan and Chinn replied that problem-based and inquiry learning "are highly scaffolded" (2007, p. 99). In tertiary education, Dochy and colleagues' 2003 meta-analysis of 43 articles on problem-based learning found a positive effect on skills (weighted effect size 0.46) and a small negative effect on knowledge that depended largely on two outlier studies.
• Workplace evidence is thin. Most studies in this debate come from schools, universities and medical education. None of the three cases measured performance on the job: the Toronto trial scored skills in a training assessment, and the Sydney and Washington studies used tests.

A large test of the discovery question is a pair of meta-analyses by Alfieri, Brooks, Aldrich and Tenenbaum, pooling 164 studies. The first compared unassisted discovery with explicit instruction.

In the authors' words, "unassisted discovery does not benefit learners, whereas feedback, worked examples, scaffolding, and elicited explanations do" (p. 1). The gap against explicit instruction was smaller for adults, coded as 18 or older (d = -0.26), than for adolescents (d = -0.53), and adults tended to benefit most from discovery tasks of both kinds. The samples were school and university learners, so read the numbers as a direction for workplace courses: give new learners guidance, and keep discovery for people who know enough to use it.

Using the three theories with an AI course builder

On 7 October 2026 I gave Coursebox a brief for the refund course: new contact-centre agents at an energy retailer, three objectives, and one learning approach per objective. The CRM field order was a behaviorist drill, with repeated practice and immediate right or wrong feedback. The refund policy was cognitivist, explained in chunks, with a worked example and a one-page rule summary. The unusual case was constructivist: a customer billed twice after moving house, followed by comparing the decision with a colleague's and reflecting on it. It is our product, so weigh my read with that in mind.

The cognitivist and constructivist parts came back as I had asked. The outline had pages called Break down the refund rules and Make a standard refund decision for the policy, and Untangle the duplicate bill and Compare decisions and reflect for the open case.

The drill did not. The CRM objective came back as a single page with no practice, and repetition is the one thing a behaviorist objective cannot do without.

The drill had landed at the start of the course, and Coursebox reserves the first page for a welcome page. In a follow-up message I asked for the drill to sit in its own section with three practice rounds. When I reloaded the outline, section 2, Enter the refund, held Practice round 1, Practice round 2 and Practice round 3, each a quiz, with 3, 4 and 5 questions. The final outline ran about 42 minutes across 5 sections and 11 pages. That confirms the structure of the drill. Whether each round's questions get harder, and whether agents log refunds correctly afterwards, needs checking in the generated course and on the phones.

Putting the three theories to work

Behaviorism, cognitivism and constructivism explain different parts of how people learn, so the useful question for a course is which one fits each objective. Lead with behaviorist drill for routines that new learners must get right every time, cognitivist explanation and worked examples for rules they must apply to new cases, and constructivist cases with coaching for problems that have no single answer, once learners know enough to reason about them. A course like the refund course needs all three, in that order for new starters.

Take one course you are building and list its objectives. Next to each one, record what learners already know and what the task asks them to do. Choose an approach, then check whether the activity and assessment support that outcome. Pilot the resulting plan and follow performance on the job. Those checks can reveal improvements and remaining gaps; attributing a change to the course requires an evaluation that considers other influences.