Measuring fish cognition: empirical‐based guidance for designing cognition assays

Interest in fish cognition has increased dramatically in recent decades and many fish have become ‘mainstream’ models for research (Patton & Braithwaite, 2015). Today, fishes are used across behavioural ecology, biomedicine, ecotoxicology and evolutionary neuroscience in tests ranging from classic maze tasks to more specialized laboratory setups and experiments conducted in the wild. Behavioural measures of cognition are crucial to understand how fishes perceive, learn, remember, integrate information to make decisions, and respond to their environment. Likewise in understanding how disease, stressors, and other factors affect these cognitve processes. Many species of fish make excellent models, with short generation times, high reproduction rates while displaying brain morphology and neural functions that are similar to those of other vertebrates (Oliveira, 2013; Triki et al., 2025). Another benefit is the diversity: as the most species rich group of vertebrates, fish together encompass a vast array of ecological niches and offer an impressive array of models to address a variety of questions (Braasch et al., 2015; Newport, 2021; Rétaux & Yamamoto, 2024; Schartl, 2014). This diversity is one of the most compelling reasons for studying fish cognition, perhaps most importantly for disentangling the evolution of cognition through comparative frameworks (Bshary et al., 2001; Bshary & Brown, 2014; Triki et al., 2025). A range of exciting questions promise that interest in fish cognition will remain high (Bshary & Triki, 2022; Healy & Patton, 2022; Miller, 2017; Salena et al., 2021; Vila Pouca & Brown, 2017). Yet, something of a sticking point is the methodological issues that researchers face when investigating cognition in fishes. Distinguishing cognitive abilities in any laboratory assay or observations in the wild ultimately depends on measuring and interpreting behavioural responses. But, it can be difficult to determine whether we are measuring the specific cognitive ability of interest (Rowe & Healy, 2014; Sarter, 2004), a stress response (Brydges et al., 2009), a response to an unintended cue we cannot even perceive. Biases due to sensory differences between researcher and animals is a common issue (Nagel, 1974), and fishes employ some relatively exotic senses, such as the mechanosensory lateral line system (Dijkgraaf, 1963) and electroreception (Schuster & Otto, 2002). In addition, there are practical challenges inherent to working with animals that live in a medium that is relatively opaque, less accessible and otherwise alien to researchers—an issue common to other aquatic animals (Nowacek et al., 2016). The enormous diversity also comes with a drawback: developing tests for vastly different species while striving for high levels of standardization and reproducibility is a challenge. This is, again, an issue common to even some of the best studied animals- such as primates (Deaner et al., 2000; Schubiger et al., 2020). Howevere, these challenges are perhaps most acute in studies of fish cognition where, apart from zebrafish and a handful of other species (Brown & Webster, 2025), most fish species have yet to mature as research models and lack well-established methodological frameworks. The development of robust and repeatable experimental methods often requires iterative testing and refinement over time (Schweinfurth & Frommen, 2025). Fundamental aspects of how to test cognition in fishes remain to be tackled. For instance, many studies rely on methods developed for other taxa, which may not be biologically or ecologically appropriate (Salena et al., 2021). Behavioural responses in tests can also be affected by basic housing or habitat conditions, with significant differences in cognitive measures observed in studies that compare levels of physical complexity (dos Santos et al., 2020; Salvanes et al., 2013; Závorka et al., 2022). These issues with task relevance and testing conditions complicate interpretation and hinder cross-species comparisons. They can also affect reproducibility (Paull et al., 2024; Webster & Rutz, 2020) and undermine the utility of null results (Schweinfurth & Frommen, 2025). An illustrative case study is the classic behavioural model species, the three-spined stickleback (Gasterosteus aculeatus). Recent findings in three areas demonstrate how methodological issues affect even historically well-studied species. Physical dimensions: Recent research has begun to reveal that sticklebacks are sensitive to differences in the physical dimensions of their containment. For example, estimates of individual behaviour depended on arena size (McInnes et al., 2025), and they discriminate between and exhibit preferences for shelter based on the dimensions of the PVC tubes provided for shelter (Jones et al., 2024). More specifically, sticklebacks tested in different configurations of a commonly used T-maze differed in level of engagement (number of times fish entered end chambers), and learning rates, with measures depending on the length of the maze arms (Jones et al., 2023). These effects of maze design may be a contributing factor to the contradictory findings in studies testing the cognitive styles hypothesis in this species: differences in the length and number of maze arms were a distinction between a study that found evidence for cognitive styles (Jones et al., 2025) and one that did not (Mamuneas et al., 2015). Social condition: The density of conspecifics in home tanks has been shown to impact behavioural responses of sticklebacks when tested alone (Jolles et al., 2016), yet social conditions in housing aquaria are largely overlooked in cognitive studies of this species. Ecological validity: A neatly designed study revealed that sticklebacks are not capable of individual recognition because fish in that study were able to associate objects as cues for food reward but did not use conspecifics as cues (Webster & Laland, 2020). However, sticklebacks social interactions are characterized by competition with conspecifics (Frommen et al., 2007; Utne-Palm & Hart, 2000), suggesting that social cues may be less relevant in feeding contexts for this species. Indeed, when testing sticklebacks in a territory defence context, a recent study demonstrated that sticklebacks could identify individual conspecifics (Sogawa et al., 2025). These examples highlight how nuanced experimental design can lead to different outcomes. While such methodological considerations are important across all scientific fields, fish cognition research often relies on single studies per species to draw broad conclusions, with less systematic exploration of how physical design elements influence results. In this special issue, we present a selection of papers that tackle key methodological challenges. The collection spans critical issues from optimal acclimation protocols and test apparatus design to social factors influencing cognitive performance. Several contributions detail the complexities of developing novel experimental setups and working with unfamiliar species, while others introduce new tracking methods aimed at improving standardization across the field. Below, we briefly summarize the contributions that deal with the search for optimal behavioural acclimation times, physical aspects of test designs, social condition and techniques for automated processes before making a case for prioritizing methods-focused publications that can guide future research in fish cognition. Ensuring that fish are acclimated and habituated to a particular setup is critical to obtaining meaningful results, especially for cognitive assays that rely on measuring in behavioural responses to novel setups. Most researchers will include some form of acclimation trials for the fish to habituate to the testing setup, but they can be time-consuming and for many species and populations there is still uncertainty about what the optimal duration and frequency of such acclimation trials should be. Varracchio et al. (2024) shed some light on a commonly studied species, guppies (Poecilia reticulata), which exhibited preferences in a choice experiment, but these preferences only became evident after 5 days of exposure to the apparatus. This study demonstrates how optimal acclimation protocols can be established for other species. Experimental setup details can significantly influence behavioural responses, with increasing evidence that distances and sizes matter. Swaney et al. (2024) describe important consequences of different distances in a classic shoaling preference test. Tasked with discriminating between familiar and unfamiliar conspecifics, zebrafish (Danio rerio) individuals showed clear preferences for familiar shoals, but only when inter shoal distances were short (30 cm); when stimulus shoals were separated by larger distances (45 or 60 cm) this preference was not seen (Swaney et al., 2024). Likewise, the dimensions of the setup used for a detour task to study inhibitory control and risk-taking in mosquitofish (Gambusia affinis) were found to be a potential reason for the inability of this species to exhibit the expected differences across experimental conditions with different levels of microplastic pollutants (Irwin et al., 2024). These studies underscore the importance of the dimensions of experimental setups when interpreting results. Dellinger et al. (2024) explore a relatively unstudied species in cognitive fields, the Arctic charr (Salvelinus alpinus). They describe several factors which hindered the ability of char to perform a spatial learning task and suggest solutions worth considering for researchers exploring other less common species. Primarily, they propose the integration of arenas into housing aquaria and acclimation times to increase the proportion of individuals that participate in trials, reinforcing previous recommendations in this vein (Salena et al., 2021). As cognitive research predominantly relies on well-established model species, these methodological insights are especially valuable for broadening taxonomic representation in the field. At a more general level, Munson and DePasquale (2024) have developed a comprehensive guide for the use of mazes in fish cognition. Building on recent work examining how maze design and assay protocols affect zebrafish performance (Benvenutti et al., 2021), their review extends these findings to other species and synthesizes critical factors to consider when designing spatial cognition assays for fishes. This resource will be invaluable for researchers developing maze-based methodologies in the future. In addition to species-level differences in cognitive abilities, variation within populations also influences assay outcomes. Individual sex, size and age can all affect behaviour, with important implications for experimental design and test arena configuration. Here Yin and Horzmann (2024) detail an empirical exploration of performance in mazes across sex and age using the classic zebrafish model. Males and older fish (more than 3 years old) learnt more slowly and made more errors than fish from younger age groups, with fish under 1 year old performing best across all metrics in a T-maze study. Exploring a more nuanced social factor of male presence and the potential of coercive mating, Ernst et al. (2024) found that there was no impact on female spatial navigation learning in porthole livebearers (Poeciliopsis gracilis). This was despite the consequences of male presence, with reduced body condition and slower growth of focal females exposed to males and the expectation and ample evidence that coercive mating impacts the female brain and behaviour in previous work. This highlights the complex and potentially species-specific nature of social influences on cognitive performance. A major area of research where fish can be useful is exploring ‘swarm intelligence’ or collective cognition. Understanding how groups coordinate and make decisions is an exciting area of research, but developing empirical assays that can unravel the mechanisms behind such decision making can be a challenge (Ioannou, 2017), for example water depth in such studies is kept shallow to better track fish, but limit behavioural responses and cause artefacts. Similarly, training protocols and appropriate cues to initiate responses vary across individuals, species and test setups. In this vein, Lecheval and Theraulaz (2024) proposed and successfully tested aversive conditioning for use in a schooling species, the rummy-nose tetra (Hemigrammus rhodostomus). They showed that they can use aversive stimuli to effectively trigger collective escape responses that propagate rapidly through small schools. A major goal for some fields is to develop high-throughput protocols and fully automate testing systems. By reducing human involvement, such approaches are aimed at minimizing biases and human observation errors, while also reducing welfare costs for fish. However, current automated tracking methods face challenges themselves. Low-cost systems that are easily adapted to multiple species or research questions are still limited. As a partial solution, Lucks et al. (2024) detail a low-cost, versatile Raspberry Pi-based system for gate opening (that removes the need for human manipulation of in-tank gates) that can be customized to fit a variety of experimental setups for testing spatial orientation and navigation. The authors illustrate the validity of their protocols with a worked example where weakly electric fish navigate through a test system. Step-by-step protocols and codes for the gate control are included for other researchers to use in their own projects. Similarly, Ajuwon, Cruz, & Monteiro (2024) describe their GoFish (Ajuwon et al., 2024) platform, a highly customizable system that automates two-alternative forced-choice test learning experiments and allows for efficient testing of larger sample sizes. Together with other tools and frameworks (Dutta et al., 2025; Giske et al., 2025), these are exciting technological advances that will have a large impact on cognitive studies of fish both in the laboratory and in the wild. While the preceding sections argue for more accurate and comparative science though standardization and careful consideration of constraints, it is important to pause and consider whether these rules are actually useful: do they help or obscure our search for answers to all questions in the field? Schuster (2024) draws on decades of experience and examples from the literature to raise the point that following methods too rigidly can cause problems. His article poses five ‘golden rules’ commonly used in cognitive and behavioural studies, and then challenges them, and us, to consider when it might be worthwhile to bend or break them. While we strive to improve our understanding of fish cognition, Schuster cautions us to bear in mind that there is no one-size-fits-all test. The search for the cognitive abilities of an animal requires experience and knowledge of that species, and sometimes substantial luck, which no amount of careful design or acclimation protocols can replace. The diversity of fish species and their natural variation in behaviour and ecology represents a fascinating source for investigating how cognitive abilities arise and how variation is maintained and affected by environmental factors. Studies of fish cognition explore a broad range of domains, from processing social information and learning (Brown & Webster, 2025; Dewenter et al., 2017; Laland et al., 2011), to how fish perceive and navigate through environments (Nava et al., 2011; Santacà et al., 2022; Sibeaux et al., 2022, 2025), and recognition (de Waal, 2019; Kohda et al., 2019, 2023; Newport et al., 2016, 2018; Vail et al., 2013) and understanding the emergence and coordination of collective behaviour (Bshary et al., 2006; Jolles et al., 2020; Munson et al., 2021) As this field expands into new species and environmental contexts, including how cognitive processes respond to anthropogenic stressors (Breedveld et al., 2025; Newport et al., 2021), establishing robust methodological foundations becomes increasingly critical. The methodological insights presented in this special issue provide essential groundwork for ensuring that our understanding of cognitive diversity keeps pace with the remarkable biological diversity we seek to comprehend. Incorporating the ecological significance of cognition is essential for understanding how cognitive abilities arise and persist. In situ observations of ‘wild cognition’ through the use of adapting methodology and technology is increasingly favoured (Bshary & Triki, 2022; Griebling et al., 2022; Mourier et al., 2025; Pritchard et al., 2016), but field studies face specific challenges that are inherent to working in a less controlled environment. A major obstacle here is tracking and identifying individual fish, which is typically more difficult than in captivity (Jungwirth et al., 2019). Here, rapid technological advances have seen the development of systems that overcome these challenges (Francisco et al., 2020). Affordable open-source Radio Frequency Identification platforms for animal identification (Bridge et al., 2019), computer vision AI tools for analysing video recordings in natural settings (Martinez-Alpiste et al., 2024; Weinstein, 2018) and innovative experimental designs for cognitive testing in the wild (Vila-Pouca et al., 2025) now offer unprecedented opportunities to study fish cognition at scale in their natural environment. The methodological considerations highlighted in this special issue, when combined with technological advances and forays into wild cognition, provide a roadmap for more robust and ecologically meaningful studies of fish cognition. Negative results and pilot studies are crucial tools for identifying issues and refining experimental methods, but often represent wasted effort when their insights remain unpublished. Pilots are essential for determining suitable cues and experimental procedures that work with fish subjects. For example, Braithwaite and Salvanes (2005) conducted several pilots to discover that using pairs of fish yielded optimal engagement because solitary fish would not interact with test stimuli at all. Despite their extensive use in laboratories, such pilots are rarely published or even mentioned in manuscripts. Similarly negative results are often unpublished because isolating any methodological reasons for the lack of expected effects often requires further work and less immediate impact. However, while there are practical difficulties in developing such studies or small pilots into full outputs, we argue that developing such work into standalone studies would be worthwhile to reduce the continual ‘reinvention of the wheel’ by making procedures and designs available to all. Similar arguments have been made before (Bespalov et al., 2019), but bear repeating. Moreover, we suggest that methodological studies of negative results and expansions of pilots provide useful avenues for guidance for research students on their first forays into relatively simple empirical projects. As a final note, we would like to thank the contributors to the issue and the many reviewers whose diligent work made this possible. The methodological insights presented in this special issue, from acclimation protocols and apparatus design to tracking innovations and species-specific considerations, represent important steps toward the goal of advancing rigorous fish cognition research. By building on these foundations and continuing to prioritize methodological transparency, the fish cognition community can ensure that our understanding of cognitive diversity keeps pace with the remarkable biological diversity we seek to understand. Nick A. R. Jones wrote the first draft, all and to the of the final not to this article as no were or the current study.