Behavioural flexibility allows animals to adjust to changes in their environment. Although the cognitive processes that explain flexibility have been relatively well studied in psychology, this is less true for animals in the wild. Here we use data collected automatically during self-administered discrimination-learning trialsfor two passerine species, and during four phases (habituation, initial learning, first reversal and second reversal) in orderto decompose sources of consistent among-individual differences in reversal learning, a commonly used measure for cognitive flexibility. First, we found that, as expected, proactive interference was significantly repeatable and had a negative effect on reversal learning, confirming that individuals with poor ability to inhibit returning to a previously rewarded feeder were also slower toreversal learn. Second, to our knowledge for the first time in a natural population, we examined how sampling of non-rewarding options post-learning affected reversal-learning performance. Sampling quantity was moderately repeatable in blue tits but not great tits; sampling bias, the variance in the proportion of visits to each non-rewarded feeder, was not repeatable for either species. Sampling behaviour did not predict variation in reversal-learning speed to any significant extent. Finally, the repeatability of reversal learning was explained almost entirely by proactive interference for blue tits; in great tits, the effects of proactive interference and sampling bias on the repeatability of reversal learning were indistinguishable. Our results highlight the value of proactive interference as a more direct measurement of cognitive flexibility and shed light on how animals respond to changes in their environment.