A contrast ratio can call two pages equal and still leave one of them harder to proofread. Black text on a white page and white text on a black page can both score 21:1. A proofreading study found the dark letters won, and the gap grew as the type got smaller.

That is the part a checker will not tell you. The number says the pair is strong. The reading test says the direction still matters, especially when the type is small.

The Same Ratio Is Not the Same Page

WCAG 2.2 asks for a contrast ratio of at least 4.5:1 for normal text, and 3:1 for large text. Large text, in that rule, is at least 18 point, or 14 point in bold. The ratio is a floor for whether text can be told apart from its background. It is not a score for which arrangement is easier to proofread.

The formula sorts the two colors by which one is lighter, then divides. It never records which color you meant as the letters. The number can match while the reading does not. Black on white and white on black are the clearest case: both hit 21:1, and they are not the same page.

Purple desk lamp with a warm orange glow floating in starry purple space

What the Proofreading Test Compared

Cosima Piepenbrock, Susanne Mayr, and Axel Buchner published the test in Human Factors. The abstract is on PubMed, and the paper is in Sage. People proofread real passages in one of two setups: black text on white, or white text on black. The advantage of the dark text on the light page grew in a straight line as the type got smaller. The authors close with a plain line: “Especially with small font sizes, negative polarity displays should be avoided.”

A write-up published October 1, 2026, and updated October 5 walks through that paper and says the passages were set at 8, 10, 12, and 14 point. Those four sizes are the useful picture. Negative polarity, in the authors’ line, means light letters on a dark page. The gap shows up as those letters shrink.

The test compared those two extremes. It did not compare two grays that both barely clear 4.5:1. It does not hand you a conversion chart for how much extra contrast a dark page needs. It does say an identical ratio is not identical proofreading.

This study does not measure eye strain, and it does not measure battery use. A dark page can still be the right look for a bar, a studio, or a night-time menu. The limit is the small light type, not the whole theme.

WCAG 2.2 Still Sets the Floor

The next version has not replaced that floor. The WCAG 3 note on text contrast still says the contrast algorithm is yet to be determined. For this draft, the requirement assumes the algorithm will include a size and weight factor. That is the working group writing down the same hole the proofreading test points at: size belongs in the rule.

Until that draft becomes the rule you are held to, 4.5:1 and 3:1 are still the numbers. Passing them is the start on a dark page. It is not the proof that a 12-point label in light gray will survive a phone in daylight.

Small Type Is Where a Dark Page Slips

Open the page a customer actually gets. On a restaurant site, that is often a dark hero with the hours, the neighborhood, and “Reserve” set in type that looked elegant on a desktop mock. Small light type on a dark page is where the gap grows.

Visual hierarchy already treats contrast as one of the levers that tells a stranger what matters. This is the finer cut of that lever. A button can be loud and the line under it — the price, the cutoff time, the street — can still be the line people misread.

What to change is ordinary:

  • Keep at least 4.5:1 for normal text. Do not treat this study as a reason to go dimmer.
  • If a line has to sit on a dark background, make it larger or heavier before you call the theme finished.
  • Check the phone, not the bright monitor in the office. Small type is where the study says the two directions pull apart.

We have a separate note on how to switch a site into a dark theme with CSS. That one is the switch. This one is whether the small text can be proofread after you flip it.

A Dark Page Still Has to Be Readable

A contrast ratio is a floor, not a reading test. Keep 4.5:1, and do not stop there when the type is small and light. If the hours, the price, or the address only work when someone pinches to zoom, the theme is ahead of the text.

That is layout work: which lines are allowed to be quiet, and which ones a stranger has to read on a phone. We do that on our UX and UI page. Bring the live URL and the line people keep missing — a price, a phone number, a booking step. Fifteen minutes on the calendar is enough to say whether the dark page needs larger type or a lighter block behind the words.