22 April 2026TAyumira Editorial

Dual Coding and Multiple Representations: The Evidence on Words Plus Diagrams

Dual coding evidence: Rexigel et al. (2024) STEM meta-analysis on multiple representations, Paivio's dual-coding theory, Mayer's multimedia principle — and classroom routines.

Dual coding is the research-backed idea that combining words with images produces better learning than words alone — but the details are where teachers get it wrong. Adding a decorative picture to a slide is not dual coding. Multiple coordinated representations that each do different work for the learner is. The 2024 Rexigel meta-analysis in STEM settled an interesting adjacent question: does going beyond two representations help, and under what conditions. This evidence review sets out what dual coding and multiple representations actually are, what the research shows, and how to use coordinated representations without overloading working memory.

What dual coding and multiple representations are

Dual coding theory — from Allan Paivio in the 1970s — proposes that information is processed through two separate channels, verbal and visual, and that encoding in both channels produces more durable learning than encoding in one. The practical implication: pair words with diagrams, graphs, or images that carry the same meaning.

Multiple representations extends the idea. A concept can be represented in words, symbols, diagrams, graphs, tables, equations, gestures, or physical models. When students learn to translate between representations — to see that a graph, a table, and an equation describe the same relationship — they develop a more connected mental model.

Richard Mayer's multimedia learning research refined the design principles. Not every word-plus-image combination helps. Coherent, contiguous, non-redundant, well-signalled combinations help. Decorative images, text placed far from the image it illustrates, and narration that competes with on-screen text each damage learning.

What the research actually shows

The evidence base is deep and well replicated.

Mayer and colleagues' multimedia learning programme has produced dozens of classroom and laboratory experiments confirming the basic dual-coding effect and the specific design principles (coherence, contiguity, signalling, redundancy, modality, personalisation, segmentation, pre-training). Effect sizes are moderate and consistent.

Rexigel and colleagues (2024) meta-analysed a specific extension question in Educational Psychology Review: does using more than two external representations in STEM produce additional benefit, and under what conditions. Their finding: small but reliable additional gains from multiple representations, strongest when the representations are coherent (show the same underlying relationship in complementary ways) and when students are explicitly taught how to translate between them. More representations do not help automatically. The coordination is what produces the effect.

The mechanism is about schema construction. When students see the same relationship expressed in words, in a graph, and in an equation, each representation reveals different features, and students build a more connected mental model. But only if they are taught to translate — the translation is the learning, not the representations themselves.

Core principles

The converged design principles from dual-coding and multiple-representation research are unusually specific.

Pair words with images that share the same meaning. Not decoration. A diagram that adds information the words do not.
Place words and images close together. The spatial-contiguity principle: text should be next to the part of the image it describes, not in a caption at the bottom or a key at the side.
Eliminate redundancy. Do not read on-screen text aloud while students are trying to process the diagram. The redundant channel loads working memory.
Signal the structure. Headings, arrows, colour coding, labels. Students need navigational cues to build the mental model.
Coordinate multiple representations explicitly. Show the graph, the table, and the equation side by side. Ask students to mark which feature of the graph corresponds to which row of the table and which term of the equation. The translation is the task.

The classroom routines

Four routines carry most of the dual-coding evidence.

Labelled diagram plus narration. Students see a labelled diagram while the teacher narrates the process. No on-screen paragraph text duplicating the narration.
Side-by-side representations with translation tasks. Graph, table, and equation of the same relationship shown together. Students annotate the correspondences.
Silent diagram re-drawing. After instruction, students re-draw the key diagram from memory, with labels. Pairs dual coding with retrieval practice.
Concept-mapping from multiple representations. Students produce a single concept map that draws on the word, graph, and equation forms they have seen.

Classroom examples across phases

Primary. Year 3 science on the water cycle. A labelled diagram with arrows shows evaporation, condensation, precipitation, and collection. The teacher narrates the diagram, pausing at each stage. Students re-draw the diagram from memory on a blank frame and label each process.

Secondary. Year 10 mathematics on linear functions. The board shows three representations side by side: a word description ("a tap fills a tank at a constant rate"), a table of values, a graph, and an equation. Students mark which feature of the graph corresponds to which term of the equation and which row of the table.

Tertiary. First-year biochemistry on enzyme kinetics. Lineweaver-Burk plot, Michaelis-Menten equation, and a schematic of enzyme-substrate binding shown together. Students annotate each representation to show how K_M and V_max appear in each, and translate between them in a structured worksheet.

Where dual coding fails

The failure modes come up constantly.

Decorative images. A stock photo of children learning is not dual coding. The image must carry meaning.
Text and image separated. Putting the diagram on slide 4 and the text on slide 5 defeats the spatial-contiguity principle. Keep them together.
Reading on-screen text aloud. The redundancy principle: if the text is on screen and also narrated, you are loading the same channel twice. Narrate a diagram, not text.
Too many representations without coordination. Six representations of the same concept is not better than three if students have not been taught to translate between them. The Rexigel meta-analysis is clear: coordination is the mechanism.
Treating dual coding as note-taking style. Students doodling in the margin is not dual coding. The doodles must be structured, meaning-carrying, and tied to the concept.

Best fit and poor fit

Best fit: STEM subjects with natural graphical and symbolic representations; any subject where a relationship or process has multiple natural expressions. Strongest recent evidence is in science, technology, engineering, and mathematics.

Poor fit: content that is genuinely word-only (literary interpretation, some philosophical argument). Adding images for their own sake to word-appropriate content damages, not helps. Dual coding is a principle applied where the content naturally supports it, not a universal prescription.

Teacher requirements, assessment, and resources

Dual coding is slide-design and worksheet-design intensive. The investment is in building coherent visuals that carry meaning — not in buying software. Free diagramming tools and hand-drawn boardwork both work; the quality sits in the design, not the medium.

Assess with translation tasks. Give students one representation and ask for the others. Give them an unlabelled diagram and ask for labels. The translation move is where the learning is tested.

How TAyumira supports dual coding and multiple representations

TAyumira generates lesson materials that apply multimedia-learning design principles. When you enable dual coding or multiple-representation support, the generator produces:

Labelled diagrams with words placed contiguously
Side-by-side multiple representations with suggested annotation tasks
Narrator notes that avoid on-screen text redundancy
A translation-task exit ticket that asks students to convert between representations
Re-drawing prompts for retrieval that combine memory and dual coding

Start for free — the Free tier covers the full workflow.

FAQ

What is the effect size of dual coding?

Mayer's multimedia learning research has produced dozens of experiments showing moderate consistent benefits of coherent word-plus-image combinations over words alone. Rexigel and colleagues (2024) reported small but reliable additional benefit from multiple coordinated representations in STEM — stronger when students are explicitly taught to translate between them.

What is the difference between dual coding and multiple representations?

Dual coding is the broader principle — pairing words with images. Multiple representations is the specific extension: using three or more coordinated expressions of the same concept (words, graphs, tables, equations, diagrams). Rexigel and colleagues (2024) showed the extension produces additional benefit when the representations are coordinated and taught explicitly.

Is dual coding the same as "learning styles"?

No. Learning-styles theory proposes that individuals have fixed style preferences (visual, auditory, kinaesthetic) and should be taught in their preferred style. The evidence does not support that. Dual coding is about combining channels for all learners — the combination is the mechanism, not matching a preferred channel.

Can images ever hurt learning?

Yes. Decorative images that do not carry meaning reduce learning — the coherence principle. On-screen text read aloud simultaneously reduces learning — the redundancy principle. Images placed far from the text they describe reduce learning — the contiguity principle. Mayer's design principles identify exactly where multimedia harms rather than helps.

How many representations are too many?

Rexigel and colleagues (2024) found small additional benefit from more than two coordinated representations in STEM, but the benefit flattens — and may reverse — if representations are not coordinated. Three to four coordinated representations is typically a defensible upper bound; beyond that, working-memory load dominates.

Related evidence reviews

Sources

Rexigel, E., et al. (2024). The more the better? A systematic review and meta-analysis of the benefits of more than two external representations in STEM education. Educational Psychology Review.
Mayer, R. E. Multimedia Learning. (Foundational programme of research on multimedia principles.)
Paivio, A. Mental Representations: A Dual Coding Approach. (Foundational dual-coding theory.)
Sweller, J. Cognitive load theory. (Foundational for understanding why the design principles matter.)

Try one dual-coding routine this week

Pick one concept in your next unit with a natural graphical expression. Build a single slide that shows word description and diagram side by side with words placed next to the parts of the diagram they describe. Narrate the diagram, not the text. If you want coherent multimedia slides generated for your topic, create a free TAyumira account.