012 min
Where Cognitive Load Theory shows up
A designer in a product team writes the first-run tutorial for a new analytics tool. The tutorial is one screen. It names six menus, explains four chart types and lists three keyboard shortcuts. The designer knows every item, so the screen looks short and complete to them. This gap between what an expert sees and what a beginner sees is the curse of knowledge.
A new user reads the screen and keeps the first two menu names in mind. By the fourth menu name, the first one is gone. The user goes back to the top, loses their place and closes the tutorial. They learned almost nothing, even though every sentence was clear.
Now change one input. The designer splits the content into three steps. Each step shows one menu, places the real interface beside the text, and asks the user to perform the action before moving on. The total content is the same. The user now finishes with a working chart.
What differs is not the amount of content. It is the number of new items the user must hold and relate at the same moment. Cognitive load theory names that quantity and gives rules for controlling it. The pattern appears wherever a person learns something new from a screen, a lecture or a document. The content can be correct and the learner can still fail, because the format asks for more than working memory can supply.
022 min
How Cognitive Load Theory works
The theory rests on two memory systems. Working memory is where conscious thinking happens. Long-term memory stores what a person already knows. Working memory is small and short-lived. Without rehearsal, almost all information held in working memory is lost within 30 seconds.
Its size is debated. George Miller's 1956 paper gave about seven items, a figure that became Miller's Law. Nelson Cowan's 2001 review argued that Miller's figure of about seven chunks was a rough estimate and not a real capacity limit. Cowan put the real limit at three to five chunks. A chunk is one unit of meaning, such as a familiar word, so the count depends on what the person already knows. See chunking.
These limits apply only to new information that has not yet been stored in long-term memory. Information that is already stored can be brought back into working memory without a known limit on how much. This is the central move of the theory. Learning is the process of storing new patterns in long-term memory, and these patterns are called schemas. A schema groups several items into one. A fluent reader sees the word "planning" as one unit. A child learning to read sees eight separate letters. The same page therefore asks very different amounts of working memory from the two readers.
What makes material hard
The theory measures difficulty by how many elements must be held and related at once. Material that contains many interacting elements is harder than material with few. Ten unrelated word translations have low interactivity, because each pair can be learned alone. Understanding how the parts of one sentence agree with each other has high interactivity, because no part makes sense alone. The theory calls this element interactivity.
Three kinds of load
The theory divides the demand on working memory into three parts.
| Kind of load | Where it comes from | Can a designer change it? |
|---|---|---|
| Intrinsic | The content: how many elements interact | Not by presentation. Only by changing what is taught at one time |
| Extraneous | The format: for example, related items shown in separate places | Yes. This is the main target |
| Germane | The learning processes that build schemas | Support it once the other two are low |
Intrinsic load comes from the content, extraneous load from how the content is presented, and germane load from the learning processes. The three add up. When the total exceeds working memory capacity, learning stops, and the only part the designer can reliably cut is the extraneous part.
032 min
Where Cognitive Load Theory comes from
John Sweller published "Cognitive load during problem solving: Effects on learning" in the journal Cognitive Science in 1988. Paas and Van Merriënboer (1994) summarise the 1988 theory as blaming high cognitive load and misdirected attention for the failure of traditional instruction. The problem in question was teaching people to solve problems in subjects such as mathematics, physics and programming.
The mechanism Sweller identified is a search process. Novices facing an ordinary problem compare the current state with the goal and look for steps that reduce the difference. This is called means-ends analysis. Paas and Van Merriënboer say this search imposes extraneous load in conventional problem solving. The search uses working memory, so the learner has little left for noticing which steps worked and storing them. A learner can solve many problems this way and still learn little.
The first experimental support for the alternative came from algebra. The worked-example effect was first demonstrated by Sweller and Cooper in 1985. They found that algebra students learned more from studying worked examples than from solving the equivalent problems. A worked example shows the full solution, step by step, so the learner does not search at all.
Sweller, van Merriënboer and Paas then published a framework paper in 1998 titled "Cognitive architecture and instructional design" in Educational Psychology Review. It linked the memory model above to a set of design effects. Chandler and Sweller had earlier tested one of those effects, split attention, in a 1991 paper in Cognition and Instruction.
041 min
What Cognitive Load Theory predicts
The theory is useful because it predicts specific effects. Each one follows from the same rule: remove load that does not help learning.
| Effect | What happens | First shown |
|---|---|---|
| Worked example | A novice who studies solved problems learns more than a novice who solves the same problems | Sweller and Cooper, 1985 |
| Split attention | Learning drops when related items sit in separate places | Chandler and Sweller, 1991 |
| Expertise reversal | A method that helps novices stops helping, then hurts, as skill grows | Kalyuga, Ayres, Chandler and Sweller, 2003 |
Split attention in detail
In the split-attention effect, domain elements that must be processed together are presented separately. Learners then have to keep one element in memory while they search for another element to relate it to. A diagram with a separate legend is the standard case. The learner reads a label, looks away to find its meaning, and holds the label in working memory during the search. Putting the text inside the diagram removes the search. The content is unchanged and the extraneous load is lower.
The separation can be in space, as in a diagram and its caption on different pages, or in time, as in a spoken explanation that arrives after the picture it describes.
051 min
A second case: when guidance stops helping
The worked-example effect has a limit that the theory itself predicts. Kirschner, Sweller and Clark (2006) state it plainly. The worked-example effect first disappears and then reverses as learners gain expertise. A learner who already knows the steps gains nothing from seeing them again, and the repeated explanation takes up working memory that the learner could have used to solve the problem.
The two cases differ in one variable: how much the learner already knows. The algebra students of 1985 were beginners. Their long-term memory held no solution patterns, so solving problems meant searching, and worked examples removed the search. An experienced learner has the patterns already. For them, solving is quick and studying an example is redundant work.
The contrast teaches something that neither case teaches alone. The same instruction can lower load for one learner and raise it for another, so "good design" cannot be judged without knowing who the learner is. This is why the theory is often applied as a rule about fading: give full worked examples to beginners, then reduce the guidance as skill grows.
062 min
How Cognitive Load Theory shows up in product and design
The theory was built for classrooms, and its ideas moved into interface design. Nielsen Norman Group defines the cognitive load of a user interface as the amount of mental resources that is required to operate the system. A user of a site must learn its navigation, layout and forms, and must also remember the facts relevant to their goal. Both draw on the same working memory.
Nielsen Norman Group applies the intrinsic and extraneous split directly. Intrinsic load is the effort of absorbing the new information the user came for. Extraneous load is processing that takes up mental resources but does not help users understand the content. Its example is font styles that do not carry any unique meaning.
Here are three situations.
- A settings form. A product designer places 25 options on one screen. A user setting up a first project must read each option, decide, and remember earlier choices. Splitting the form into a short sequence, with sensible defaults, shows only the few options that matter first. This is the idea behind progressive disclosure.
- Documentation for a library. A technical writer puts the explanation of a function in one section and its code example in another, five screens apart. The reader must hold the explanation while scrolling to the example. Placing the example directly after the explanation removes the search.
- A data dashboard. An analyst reviewing model results sees twelve charts with different colour schemes. Each new colour scheme has to be decoded before the data can be read. One shared scheme leaves the effort for the data.
In each case the designer cut extraneous load. None of them removed content the user needed.
071 min
How to use Cognitive Load Theory on purpose
The theory gives a short list of actions. Each works by lowering extraneous load or by matching guidance to what the learner knows.
Show worked examples to beginners.
Give a complete solved case before asking for a solution. This removes the search that fills working memory.
Put related items together.
Place labels inside diagrams, code next to its explanation, and the error message next to the field that caused it.
Remove what does not help the task.
Nielsen Norman Group names redundant links, irrelevant images and meaningless typography as sources of delay.
Reduce what the user must remember.
Look for anything in your design that requires users to read text, remember information, or make a decision. Replace it with a visible value, a re-displayed earlier answer or a default.
Fade guidance as skill grows.
Offer a beginner path and an expert path. Experts then skip the steps they no longer need.
Measuring load
Load is hard to observe directly. Paas (1992) separated mental load, which the task and instructional design impose, from mental effort, which is the capacity a learner actually allocates. Studies since then often ask learners to rate their effort after a task. Treat such ratings as a signal to compare two designs, not as a precise measure of load.
082 min
Where the evidence is contested
The theory has critics, and the strongest criticism concerns germane load. Ton de Jong (2010) reviewed the theory and noted that cognitive load theory has recently been the subject of criticism regarding its conceptual clarity. His own objection is specific. The usual definition says that load from relevant mental activity is germane and helps learning, and load from interfering activity is extraneous and hurts it. De Jong says the first issue apparent in this definition is its quite tautological character: what is good is good, and what is bad is bad.
A second objection is that the three loads are defined differently. Intrinsic load describes the material, and germane load describes what the learner does. De Jong asks how a property of the material and a process in the learner can be added together. A third objection is that the same activity may count as germane for a novice and as extraneous for an expert, so the label depends on the learner and cannot be read from the design alone.
Here is what this means for a practitioner. The worked-example effect and the split-attention effect rest on controlled experiments and are used widely. The three-way split is less secure, and the claim that designers should raise germane load is the least secure part of it. Until the definitions are settled, measure the effect of a design change on learner results, and do not rely on a classification of load types as proof.
091 min
Common misunderstandings
"Less cognitive load is always better." Nielsen Norman Group states that there is no way to eliminate cognitive load entirely, and that even if this was possible, it would not be desirable. People visit a site to learn something they did not know, and absorbing it is the point. Only the extraneous part should be cut.
"People can hold seven things in mind." This is the popular reading of Miller. As the review by Cowan above notes, seven was a rough estimate and the more defensible limit is three to five chunks. The number also depends on chunking, so any single figure is a rough guide.
"The theory says to give learners less to do." Kirschner, Sweller and Clark argue for strong guidance for novices because they lack the stored knowledge to search a problem efficiently. The same paper says guidance can be relaxed as expertise grows. The theory does not prefer easy tasks. It prefers tasks in which working memory is spent on learning.
"It applies to every interface." The theory is about learning new material. A routine task that the user has done many times draws on stored schemas and loads working memory far less. The ideas still help, but expert users of a familiar tool are a different case from first-time users.
101 min
Cognitive Load Theory vs. nearby concepts
The names in this area are close, and the concepts answer different questions.
| Concept | What it is | How it differs from the theory |
|---|---|---|
| PsychologyWorking memory | ||
| Not in the library yetCognitive load | ||
| PsychologyDual coding | ||
| DesignHick's Law |
The axis that separates the theory from all of these is its aim. It does not describe memory for its own sake. It asks how to arrange instruction so that a limited working memory is spent on learning.
?7 questions
Questions people ask
What is an example of cognitive load theory in design?
What are the three types of cognitive load?
What is the difference between cognitive load and cognitive load theory?
How do you reduce extraneous cognitive load?
What is the split-attention effect?
What is the expertise reversal effect?
Is cognitive load theory still valid?
§7 sources
Sources and further reading
Show all 7 sourcesHide sources
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257-285.
Sweller, J., van Merriënboer, J. J. G., & Paas, F. G. W. C. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10, 251-296.
Paas, F. G. W. C., & Van Merriënboer, J. J. G. (1994). Instructional control of cognitive load in the training of complex cognitive tasks. Educational Psychology Review, 6(4), 351-371.
Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1).
Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why minimal guidance during instruction does not work. Educational Psychologist, 41(2), 75-86.
de Jong, T. (2010). Cognitive load theory, educational research, and instructional design: some food for thought. Instructional Science, 38, 105-134.
Whitenton, K. (2013). Minimize cognitive load to maximize usability. Nielsen Norman Group (a practitioner source).





