The bottleneck the theory names
Working memory is small. Most people can hold around four chunks of information at once and process only a few at the same time — a constraint Alan Baddeley’s model of working memory maps in detail. John Sweller, a psychologist at the University of New South Wales, built cognitive load theory on exactly that bottleneck. His core claim, developed through the 1980s and 1990s, is that the design of an explanation determines how much of that limited capacity gets used for actually learning versus for struggling with the presentation.
Sweller distinguished three kinds of load. Intrinsic load is irreducible: it is the genuine complexity of the material, determined by how many elements must be understood in relation to each other simultaneously. Extraneous load is the waste — effort burned on a poorly designed diagram, a split-attention layout, an unnecessarily cluttered interface, or a worked example that hides its own logic. Germane load was originally framed as productive effort spent building schemas, though subsequent theorists have questioned whether it is meaningfully separable from intrinsic load rather than simply the part of intrinsic load that gets successfully processed.
The practically important distinction is the first two: genuine complexity versus avoidable friction.
What the evidence shows
The clearest demonstrations have come from instructional design experiments. The split-attention effect is among the most replicated: when a diagram and its explanatory text are placed apart, learners must hold one in working memory while visually searching for the other, and performance drops measurably compared to an integrated version.
The worked-example effect is similarly robust in early skill acquisition — novices who study worked solutions learn more efficiently than those who attempt equivalent problems from scratch, because problem-solving under uncertainty imposes search costs that eat into the capacity available for learning the underlying structure.
The underlying logic is the same in every case: when the form of a communication adds processing demands over and above those the content genuinely requires, understanding suffers.
Both effects have survived considerable scrutiny. A 2019 meta-analysis drawing on decades of instructional research found consistent support across domains for the proposition that reducing extraneous load improves performance, particularly for novices. The expertise reversal effect sits alongside this as a necessary qualification: support structures that reduce load for a beginner become redundant — and then actively harmful — as expertise develops, because the expert’s richer schema handles the material efficiently without scaffolding.
Where the theory breaks down
The theory has not escaped criticism. Some researchers argue that the three-load taxonomy is difficult to operationalise independently; you cannot always measure intrinsic, extraneous and germane load separately, which makes the framework better as a design heuristic than as a precisely testable mechanistic account. Others note that most experimental work has used relatively simple materials over short time spans in controlled conditions, and the question of how consistently the effects scale to complex, authentic learning over weeks remains open.
Why it matters beyond instructional design
Cognitive load theory is often read as a story about classroom materials, but its implications reach anywhere explanation happens — software interfaces, medical consent forms, legal documents, data visualisations. This is not a novel intuition; what Sweller’s framework supplies is a mechanism — the working-memory bottleneck — and a testable vocabulary for diagnosing where the waste occurs.
The practical heuristic the theory generates is conservative rather than revolutionary: remove what does not need to be there. Redundant text alongside a self-explanatory diagram, decorative graphics that share space with instructional content, and explanations that force the reader to construct connections the author could have made explicit — all of these are extraneous load in operational clothing. The evidence that eliminating them helps is among the more replicable findings in applied cognitive psychology, which, given how uneven replication has proved across the field, is itself worth noting.