Measures of cognitive performance, derived from psychometric tasks, have yielded important insights into the factors governing cognitive variation. However, concerns remain over the robustness of these measures, which may be susceptible to non-cognitive factors such as motivation and persistence. Efforts to quantify short-term repeatability of cognitive performance have gone some way to address this, but crucially the long-term repeatability of cognitive performance has been largely overlooked. Quantifying the long-term repeatability of cognitive performance provides the opportunity to determine the stability of cognitive phenotypes and the potential for selection to act on them. To this end, we quantified long-term repeatability of cognitive performance in wild Australian magpies over a three-year period. Cognitive performance was repeatable in two out of four cognitive tasks—associative learning and reversal-learning performance was repeatable, but spatial memory and inhibitory control performance, although trending toward significance, was not. Measures of general cognitive performance, obtained from principal components analyses carried out on each cognitive test battery, were highly repeatable. Together, these findings provide evidence that at least some cognitive phenotypes are stable, which in turn has important implications for our understanding of cognitive evolution.