Do Goldfish Really Have a Three-Second Memory?
"Goldfish have a three-second memory" gets used as a punchline, a metaphor for short attention spans, and sometimes as an actual excuse for keeping a fish in a bare bowl. Fact-check pages debunking it are everywhere, and they all cite the same handful of studies. We went looking for those studies by name.
The claim
Goldfish (and, by extension, aquarium fish generally) can only retain information for about three seconds before forgetting it, making anything beyond immediate stimulus-response impossible for them to learn or recall.
Source trail
Nobody we found — not the two mid-size explainers, not the four small independent pages holding page-one search positions for this question — traces the specific figure "three seconds" to an actual study, researcher, or date. It's treated uniformly as pre-existing folklore, most often blamed on people keeping fish in small, neglected bowls. What gets offered as the counter-evidence is a short, repeated list: "a 1966 University of Michigan study," a 2022 experiment where goldfish steered a wheeled robot on dry land, a 2022 Oxford-linked distance-estimation study, and a claim that a fish trained with a sound was released into open water and remembered it five months later. We checked what we could reach directly.
What's confirmed
- The 1966 study is real, has a real citation, and none of the pages repeating it name that citation. Confirmed directly via PubMed Central (checked 2026-08-18): "Stages of memory formation in goldfish: evidence for an environmental trigger," R. E. Davis and B. W. Agranoff, Proceedings of the National Academy of Sciences, volume 55, issue 3, pages 555–559, March 1966. Two mid-size explainers we checked directly, IFLScience and Discover Magazine (both checked 2026-08-18), both call it "a 1966 University of Michigan study" with no author names and no journal. Two small independent pages holding page-one positions for this query, thegoldfishtank.com and English Plus Podcast (both checked 2026-08-18), go further and name no specific study at all — the first says only "at least two separate studies," the second only "research has shown."
- A 2022 study shows goldfish learning a multi-day task in an environment they never evolved for. Confirmed via EurekAlert (checked 2026-08-18), the AAAS's own press-release wire, republishing Ben-Gurion University of the Negev's release: six goldfish were trained to steer a wheeled "Fish Operated Vehicle" toward a target on dry land, and "after a few days of training, the fish navigated to the target," recovering from interruptions and ignoring decoy targets. The release states the paper was published in Behavioural Brain Research, DOI 10.1016/j.bbr.2021.113711 (2022) — we could not load the journal article itself directly this session (see below), so this finding is sourced to the university's own press release, not independently re-derived from the paper.
- The specific "released into open water, remembered a sound five months later" story does not match the research it supposedly comes from. Checked directly via the ABC (Australia) Science Archive, published 19 August 2004 (checked 2026-08-18): the article names two distinct pieces of research — Jonathan Lovell of Plymouth University's Institute of Marine Studies trained (unspecified) fish to swim toward a sound, and Culum Brown documented crimson spotted rainbow fish — not goldfish — remembering to avoid a trawling net for up to 11 months. Neither one, as the article actually describes it, is "a goldfish released into open water that came back to a sound after five months." The popular version appears to merge the two into a story that matches neither.
What isn't confirmed
-
We could not independently read the 1966 paper's own experimental description.
Both
pnas.organdscience.org(for a related 1966 paper on the same research, "Actinomycin D Blocks Formation of Memory of Shock-Avoidance in Goldfish," Agranoff, Davis and Brink) returned HTTP 403 to direct fetch this session, and PubMed Central serves the 1966 issue only as scanned page images, not extractable text. The specific experimental details reported by IFLScience and Discover Magazine — 20 shock-avoidance trials, a retraining session three days later, memory blocked by a puromycin injection — are their paraphrase of the paper, not something we independently verified against the primary text ourselves. - The "five months" figure does not describe any study we could locate. Neither Lovell's nor Brown's research, as reported by the one source that names both, matches the commonly circulated version. We could not find an original source for the five-month figure at all — it may be a further garbling of Brown's 11-month rainbow-fish figure, but that is a guess, not a finding, and is reported here as one.
-
A third cited study, on distance estimation, is named but not independently checked.
Both IFLScience and Discover Magazine cite a 2022 paper (Sibeaux et al.) in Proceedings of the
Royal Society B, DOI 10.1098/rspb.2022.1220, on fish estimating distance using visual cues.
royalsocietypublishing.orgreturned HTTP 403 to direct fetch this session, so its existence rests on two independent secondary sources agreeing on the same DOI, not on us reading it ourselves. - Where "three seconds" itself first came from remains untraced. Every source we checked, including the ones most careful about citing the counter-evidence, treats the number as ambient folklore with no identifiable first appearance — not a gap we could close this session.
Verdict
The myth is false, and that much holds up on real evidence: a 1966 peer-reviewed study that most debunking pages cite by year but strip of its authors and journal, and a 2022 study showing goldfish learning a genuinely complex multi-day task in an environment nothing like the one they evolved for. Neither fits inside a three-second window. But one specific, often-repeated piece of supporting evidence — the fish trained with a sound, released, and remembering it five months later — traces back to two real but different studies, about a different species doing a different thing, merged into an anecdote that matches neither. The three-second claim is wrong; not everything offered to prove it wrong is right either.