Fish Cognition and Memory: What the Research Shows
Peer-reviewed studies on fish memory, conditioning, tool use in wrasses and face recognition, each tied to the species tested, and why the three-second memory claim does not hold.
Overview
Fish cognition is an active research field covering spatial navigation, numeracy, learning and decision-making. A 2021 review of the field (Salena et al., Animal Cognition) noted that most of this work is done in laboratories with captive-bred fish, so results from one species under controlled conditions cannot be assumed to apply to every species or to a home aquarium. The findings below are therefore given with the species that was actually studied.
The three-second memory myth
The popular claim that fish, and goldfish in particular, forget everything within a few seconds is contradicted by experimental results summarised in the Wikipedia article on fish intelligence. Reported examples include:
- Goldfish (Carassius auratus) remembered the colour of a tube that dispensed food one year after the tube was last presented.
- Channel catfish responded to a human voice call that had announced feeding five years after last hearing it.
- Red Sea clownfish recognised their mate 30 days after the mate was experimentally removed.
- Rainbow trout remembered how to press a bar for food three months after last seeing the bar.
- In the cichlid Neolamprologus pulcher, about 80% of fish trained on a disc-lifting foraging task still remembered it after 12 days without reinforcement, and 55% after 48 days (Costiuc, Latchem & Balshine, 2025, Animal Cognition).
Laboratory work on zebrafish (Danio rerio) points the same way. Cavallino and colleagues (2026, Frontiers in Veterinary Science) used a long-term memory paradigm in which pairs of male zebrafish met twice; fish showed behavioural evidence of remembering their earlier opponent, and that memory was lost in fish given an amnesic agent after the first encounter.
Learning and conditioning
Associative conditioning is the standard tool for testing what fish perceive and remember. In a 2025 study (Patel et al., Animals), adult zebrafish learned to associate one complex sound with a rewarding video of a zebrafish shoal and another with an aversive image; after twelve training runs they moved towards or away from the screen according to the sound, and the difference largely persisted on the second day of testing, which the authors interpret as overnight memory consolidation.
Goldfish trained in a two-choice task to distinguish plastic models of a frog and a turtle still chose correctly when the objects were rotated, performing above chance at all tested angles; recognition was easier for coloured models than for models painted black (DeLong et al., 2026, Learning & Behavior). Wikipedia also cites a 2020 study that recorded neural activity in freely swimming goldfish and found evidence that they form navigational cognitive maps.
Tool use in wrasses
Several wrasses (family Labridae) use hard surfaces as anvils to open shelled prey. The orange-dotted tuskfish (Choerodon anchorago) has been described fanning sand to uncover a bivalve, carrying it several metres and smashing it against a rock; similar anvil use is reported for the sixbar wrasse (Thalassoma hardwicke). In New Caledonia, the graphic tuskfish (Choerodon graphicus) was filmed making up to 28 strikes on two anvils within six minutes (Pryor & Milton, 2019), and a later survey recorded 16 tool-use events in 10 hours, with rocks, a shell, dead coral and a concrete mooring used as anvils and only molluscs carried to them (Pryor, 2020).
A comparative brain study (Estienne et al., 2024, Communications Biology) found that tool-using wrasses have an enlarged telencephalon and ventral forebrain and midbrain compared with other teleosts, but the difference comes from large fibre tracts linking the pallium with the inferior lobe rather than from a higher proportion of neurons.
Recognising faces, individuals and humans
A 2024 review (Kohda, Sogawa & Sowersby, Frontiers in Psychology) lists evidence of face-based individual recognition in 10 fish species from 7 families, including the daffodil cichlid (Neolamprologus pulcher), the three-spined stickleback (Gasterosteus aculeatus), medaka (Oryzias latipes) and the guppy (Poecilia reticulata). Daffodil cichlids distinguished familiar from unfamiliar faces in under 0.5 seconds, and medaka and daffodil cichlids both show a face inversion effect similar to that seen in primates.
Recognition of humans has been tested directly in the archerfish Toxotes chatareus. Using a two-alternative forced-choice procedure, Newport and colleagues (2016, Scientific Reports) trained archerfish to discriminate among 44 human face images, and they still succeeded with 18 faces after colour, head shape and brightness cues were controlled. No fetched study tested whether ordinary aquarium fish recognise their particular keeper, so that question remains open in these sources.
Mirror self-recognition and the debate
The bluestreak cleaner wrasse (Labroides dimidiatus) has passed the mirror mark test: fish given a parasite-like mark on the throat scraped it against the substrate after viewing themselves in a mirror. A 2025 study (Sogawa et al., Scientific Reports) reported that mirror-naive cleaner wrasse reached self-recognition rapidly. The interpretation is disputed: Gordon Gallup argued that the results are better explained by an instinct to remove parasites, and the 2024 review acknowledges the alternative that the fish may treat its reflection as a familiar individual rather than as itself.
Habitat and cognition
A 2026 narrative review (Klinke & Brown, Animal Cognition) reports that fish from structurally complex habitats often have larger brains and larger regions such as the telencephalon and cerebellum, and that barren hatchery rearing can produce smaller brains, an effect that physical enrichment can reduce. This link is covered in the aquarium-environmental-enrichment-guide; welfare aspects are discussed in the fish-pain-and-welfare-science-guide and stress-in-fish-guide.