Cognition arguably drives most behaviours in animals, but whether and why individuals in the wild vary consistently in their cognitive performance is scarcely known, especially under mixed-species scenarios. One reason for this is that quantifying the relative importance of individual, contextual, ecological and social factors remains a major challenge. We examined how many of these factors, and sources of bias, affected participation and performance, in an initial discrimination learning experiment and two reversal learning experiments during self-administered trials in a population of great tits and blue tits. Individuals were randomly allocated to different rewarding feeders within an array. Participation was high and only weakly affected by age and species. In the initial learning experiment, great tits learned faster than blue tits. Great tits also showed greater consistency in performance across two reversal learning experiments. Individuals assigned to the feeders on the edge of the array learned faster. More errors were made on feeders neighbouring the rewarded feeder and on feeders that had been rewarded in the previous experiment. Our estimates of learning consistency were unaffected by multiple factors, suggesting that, even though there was some influence of these factors on performance, we obtained a robust measure of discrimination learning in the wild.