The Mantis Shrimp Sees Fewer Colors Than You

The infographic is everywhere. A cartoon mantis shrimp, eyes radiating a sixteen-band rainbow, captioned with some version of "it sees colors you can't imagine." Matthew Inman's 2012 Oatmeal comic put the idea on the cultural map [S5], and more than a decade later the meme is still self-perpetuating — rainbow-eyed shrimp on T-shirts, in TED talks, in chemistry-class slide decks about how limited human vision supposedly is.

The factoid is wrong in a particularly clean and instructive way. The mantis shrimp does not see more colors than you. It sees fewer, by roughly an order of magnitude [S1], and the reason rewrites what "better" eyes even means.

In 2014, Hanne Thoen and colleagues at the University of Queensland ran the controlled experiment nobody had quite gotten around to running [S1]. They trained Haptosquilla trispinosa to associate a fiber-optic wavelength — 470 nanometers, a clean blue — with a food reward, then offered the trained color paired with a distractor wavelength some distance away in the spectrum [S1]. At 50 to 100 nm of separation, the shrimp picked correctly. At 12 to 25 nm, performance collapsed to chance: the animals chose at random [S1]. A trichromatic human, by comparison, reliably discriminates wavelengths 1 to 4 nm apart [S1]. The animal with sixteen receptor classes is roughly ten times worse at the task than the animal with three.

A note on that sixteen: the popular figure conflates color receptors with polarization receptors. The mantis shrimp midband actually carries twelve narrow-band visible-and-UV photoreceptor classes spanning roughly 300 to 720 nanometers; the remaining channels see polarization, which is something else entirely [S2].

To see why twelve receptors lose to three, consider what your own visual system is doing right now. The three cone types — long, medium, short — have broadly overlapping response curves; none of them is a "red detector." A yellow photon excites the long cone strongly and the medium cone almost as much, while a green photon flips that ratio. Your retina and visual cortex never read absolute signals. They compute differences between cone outputs and pass opponent ratios up the chain [S3]. The information lives in the comparison, not in the channel. Three broadband sensors plus a brain that takes their ratios can interpolate roughly a million distinguishable colors [S3].

Now consider what the mantis shrimp appears to be doing instead. Twelve receptor classes, each tuned to a narrow band, each largely non-overlapping with the next. Thoen's interpretation, argued from the behavioral data, is that the eye performs no opponent comparison at all [S1]. The midband sweeps across a target and each receptor fires a local yes/no signal — red-ish or not-red-ish, blue-ish or not-blue-ish — like a barcode scanner walking across a label [S1]. Recognition happens in the eye stalk. The central nervous system never computes color the way your visual cortex does [S1].

That is hardware-level identification with no software cost. It is also the trade you would design if you were optimizing for speed at almost any price.

Speed matters here more than almost anywhere else in the animal kingdom. The peacock mantis shrimp's hammer strike, measured by Sheila Patek's lab at Duke, reaches a peak velocity of about 23 meters per second from a standing start in under 800 microseconds, with peak accelerations exceeding 10,000 g [S4]. The whole see-identify-fire sequence runs in single-digit milliseconds. A cortical color-opponent computation — the math your brain is performing to read this sentence — takes too long. The shrimp got a discount on its visual system by skipping the math.

That is the clean version of the story, and it is the version the meme deserves. The actual scientific situation is messier, and the messiness is interesting in its own right. Zaidi, Marshall, Thoen, and Conway followed up in 2014 with a reanalysis arguing the behavioral data do not strictly rule out an opponent-coding system; they only show that if one exists, it performs poorly in the visible band [S3]. Bok and colleagues then demonstrated that mantis shrimp discriminate ultraviolet wavelengths much more finely than visible ones, reliably telling 351 nm from 379 nm — a 28-nanometer gap they botch in blue [S2]. A 2022 review from the Marshall lab itself is titled Colour vision in stomatopod crustaceans: more questions than answers [S2]. The barcode-scanner model is a leading hypothesis, not a settled finding.

What survives every reanalysis: the shrimp is dramatically worse at visible-band color discrimination than a creature with a third its number of photoreceptors [S1][S2]. The factoid that built the meme — sixteen receptors, superhuman vision — is upside down.

The generalization hiding underneath is the part worth keeping. Every popular-science line of the form "animal X has Y times more receptors than humans" smuggles in the assumption that more sensors equal better perception. That assumption is almost always wrong. Perception is not a sensor count; it is a sensor count multiplied by however much downstream computation an organism can afford to run on the signal. Trichromat humans see more colors with fewer cones because we paid for a cortex.

The mantis shrimp didn't, and didn't need to. Its eye recognizes prey the way a checkout scanner recognizes a can of soup — not by understanding the label, but by matching a pattern fast enough that the hammer is already moving before the question of what was seen has finished being asked.