The Law of Similarity
When things can’t sit closer together, the eye groups them another way — by what they have in common.
Module 1 grouped by distance. But distance runs out — a legend scattered across a chart, rows locked to a grid, links buried in a paragraph. When you can’t move things closer, the eye reaches for its second rule: likeness. Elements that share a colour, a shape or a size read as one set, however far apart they sit.
This is the law that lets a design speak in categories — every link the same blue, every error the same red, every primary button the same shape. Get it right and a stranger reads your groups without a word of instruction. Get it wrong and unrelated things fuse into one, or a lone oddity screams for attention it never earned. You’ll feel all of it in three quick labs.
- Group scattered elements by a shared trait — colour, shape, size — with no change in spacing.
- Feel why a single different item “pops out” instantly — and why a combination doesn’t.
- Use consistency as a signal: same role, same look.
- Code categories so people scan them at a glance — without excluding the colour-blind.
Nothing moved. Groups appeared.
In Module 1 you made groups by moving dots closer. Here the spacing is locked — a perfectly even five-by-five grid, every dot the same distance from its neighbours. Proximity has nothing left to say.
So watch what happens when you change one thing that isn’t distance. Tint a few columns, or square off a few rows, and the grid instantly splits into groups — columns, or rows — without a single dot moving. That’s similarity: the eye ties together whatever looks alike.
Count the urgent ones
Here’s a to-do list with three priorities mixed together. Your job: find how many are P1 — urgent. In the first state, priority is buried in a tiny code at the end of each row, so you have to read all eight to be sure.
Now colour-code the tags. The three urgent tasks light up in the same orange — and suddenly you count them in a single glance, without reading a word. That’s similarity turning a reading task into a seeing one.
Find the odd one
Similarity is so fast it feels like magic — but only for one trait at a time. Below is a field of shapes with a single odd one hiding in it. Hit start, find it, and watch the clock.
In Single feature, the odd one differs by colour alone — it pops out in a heartbeat, no matter how many others there are. Switch to Conjunction, where the target shares its colour with some and its shape with the rest, and the magic dies: you have to hunt item by item. That gap is the whole truth about similarity.
Consistency is a promise the eye can read
Every lab here is one law wearing different clothes. Name it, and “be consistent” stops being a chore and becomes a tool.
Same role, same look. Different role, different look.
Similarity is the eye’s second grouping rule, and it works on a small set of visual variables — colour, size, shape, orientation, texture. Colour is the loudest; the others follow. That hands you a rule you can apply to any interface: give the same job the same look, and a different job a visibly different one. Do that and your design starts speaking in categories the moment someone glances at it.
This is what “consistency” actually buys — not neatness, but grouping. Every link that shares one style becomes a set the eye collects at once; every primary button that matches becomes a single, learnable thing. The two failure modes are opposite. Too little similarity, and a lone off-style element looks like it does something special when it doesn’t. Too much, and everything carries equal weight, so nothing leads and hierarchy dies. Sameness and difference are both tools; spend each on purpose.
One caution the labs can’t show you: the strongest variable, colour, is the one roughly one in twelve men can’t fully see. So code by colour and a second cue — a shape, an icon, a label. Redundant coding groups for everyone. The working test is brutal and quick: squint until detail blurs, or drain the colour from the screen — do your groups still hold together? If they vanish, similarity was carrying more than it safely can.
Which one groups better?
Five face-offs. Same content, two spacings — one makes the relationship obvious, one makes you guess. Pick the clearer, then see why.
Grouping by likeness, everywhere
Once you can see similarity, you catch it sorting the world into kinds — on every screen, and far beyond them. Five places you lean on without a thought.
Blue and underlined, wherever it lands. You spot every clickable word on a page at once, because they all share the same look — a group scattered through the text.
The bars are scattered along the axis, but each colour belongs to one series. The legend and the bars group across the whole chart by nothing but a shared hue.
Green is always “good,” red always “wrong,” amber always “waiting.” Run your eye down a list and the reds group themselves — the problems find you.
Keywords one colour, numbers another, function names a third. You read code faster because each kind of token shares a hue — similarity sorting the text into types as you scan.
Work is blue, health is green, personal is red — across two days and six events, you see your week’s shape by colour alone, long before you read a single title.
Proximity and similarity both group things that are fully there. But the eye will also group things that are not there — it completes a broken circle, closes a dotted outline, reads a shape from a handful of gaps. When perception fills in what’s missing, that is closure, and it’s Module 3.
The real references
The Gestalt psychologists named this law a century ago, and the field has tested it ever since. Five worth your time. (Opens in a new tab.)
The Similarity glossary
Six words that turn “it looks inconsistent” into a precise, fixable diagnosis. Tap a card to flip it.
Questions you actually ask
What people get wrong
Read the grouping
Ten questions in seven formats, tagged by thinking level — each with layered hints: a nudge, the reasoning, then a deeper connection. A wrong answer opens a door, never a dead end. Served from the question bank, never hardcoded.