The goldfish three-second memory, and the studies that were run instead. · published 3 October 2026
The three-second figure has no study behind it at all; when goldfish were given a lever that released food for only one hour in twenty-four, they learned to press it around that hour, and rainbowfish trained to escape a model trawl still escaped quickly eleven months later.
These are the working notes, unedited. Everything the video
states as fact comes from here, and nothing else may be stated as fact.
The sections on limits and on what the finding does not show are
part of the brief, not an afterthought — see
the method.
The claim. Goldfish are said to have a memory of three seconds, sometimes seven. The figure is repeated everywhere and attached to no study. There is no original experiment to cite, because there was never one: the number is folklore, and it has long been convenient folklore, since an animal that forgets its surroundings every few seconds cannot mind living in a bowl.
Note what that means for the shape of this. There is no finding to overturn. The honest position is that a widely repeated claim has no evidence behind it, and that the experiments actually run point the other way.
The goldfish study. Philip Gee, David Stephenson and Donald E. Wright, “Temporal discrimination learning of operant feeding in goldfish (Carassius auratus)”, Journal of the Experimental Analysis of Behavior, volume 62, issue 1, 1994, pages 1-13. Eight goldfish were trained to work a lever that released food when pressed. The window in which the lever did anything was then narrowed, step by step, until it worked for just one hour in every twenty-four.
What they found. The fish concentrated their lever pressing around the hour when food was actually available, rather than pressing at random through the day. They were anticipating the feeding time. Whatever a goldfish is doing there, it is carrying information across many hours and acting on it.
The longer-memory study, in a different fish. Culum Brown, “Familiarity with the test environment improves escape responses in the crimson spotted rainbowfish, Melanotaenia duboulayi”, Animal Cognition, volume 4, issue 2, 2001, pages 109-113. Rainbowfish faced with a model trawl net learned over five runs to find the single hole in it and escape faster. Eleven months later they were shown the trawl again, and their escape times were still short.
This is the result usually meant when people say fish remember for months. It is worth being exact about it: those were rainbowfish, not goldfish. It does not transfer automatically between species.
Limits
The goldfish work is eight animals in a laboratory operant task, and what it demonstrates is anticipation across a daily cycle - hours, firmly, not months. The rainbowfish result is a different species, small, and measures retention as a faster escape rather than anything one could call recollection. Neither study was designed to measure the outer limit of a fish’s memory, and “no source exists for the three-second claim” is an absence of evidence, which is a weaker thing than a measurement.
What this does not show
It is not a claim that goldfish think like mammals, or that any particular tank size is or is not adequate. The welfare argument about bowls is a real argument, but it rests on oxygen, water volume, filtration and swimming space, and it neither needs this evidence nor is settled by it. The narrow point is only that the three-second figure is invented, and that when anyone bothered to test it, the fish remembered.