Elements that look alike are perceived as one category

What it is

Give a set of elements the same value on some visual attribute — the same color, shape, size, texture, or orientation — and they get read as belonging to one category, even scattered across the layout with no spatial adjacency. This cue is similarity.

It answers a different question than proximity does: proximity works off a continuous quantity, spatial distance; similarity works off whether a discrete attribute takes the same value, with distance playing no part. Two icons far apart on a page but sharing a color still get linked by similarity — the link doesn't need them to sit next to each other.

Why it happens

Similarity grouping runs on a handful of relatively independent feature pathways in the visual system — color, orientation, size, and shape each have dedicated early processing that extracts "what value does this element carry on this channel." When a set of elements share the same value on one channel, that channel tags them as one class, and the tagging happens before the elements' identity or meaning is recognized.

Because each channel is computed separately, similarity grouping is usually most solid along a single attribute: elements alike in color but different in shape get sorted mainly by color; elements alike in shape but different in color get sorted mainly by shape. Change several attributes at once and inconsistently, and grouping gets less certain, because the different channels are handing back conflicting answers.

Studying it

Similarity grouping is commonly measured with texture-segregation-style paradigms: embed a subset of elements differing on one attribute (orientation, color, size) within a field of background elements, and check whether that subset pops out automatically and quickly, scored by detection time or accuracy as a measure of that attribute's grouping strength. A classic finding from this line of work is that some attributes (color, a large orientation difference) support segmentation that needs almost no item-by-item search, while other attribute combinations require comparing elements one at a time.

In interface research this method is typically used to check whether a visual code — using color to mark status, shape to mark type — actually lets users classify items without reading each one individually.

Where it stops holding

Similarity produces a perceptual judgment of "looks like the same category"; it doesn't guarantee the elements actually belong together functionally or semantically — that gap is its main practical risk, and its specific consequences are covered separately within this group.

When strong spatial clustering is also present (a large spacing difference), similarity judgments tend to give way to the spatial cue; similarity is better suited to elements that are physically scattered and can't be grouped by position.

Applying it

• To let users sort a set of elements without reading each item, pick one visually distinct attribute (color or shape, generally not both at once) rather than varying several attributes inconsistently, which makes the grouping read as noisy.
• Elements of the same category don't need to sit together to read as one group — a consistent color or graphic language can visually link them across scattered locations on the page.
• To verify, shrink the layout until only color blocks and outlines are legible (blur it, or view it from a distance) and check whether the target category can still be picked out quickly without reading any text.

Related

• Same group: A2.02.2 inferring sameness from similarity causes functional misjudgment · A2.02.3 differently-functioning elements must be visually distinguishable
• Nearby: A2.01 proximity · A2.06 figure-ground · A2.10 conflict and priority among Gestalt principles
• Search terms: similarity · visual search · texture segregation · perceptual grouping