Objective Physicians’ cognitive empathy is associated with improved diagnosis and better patient outcomes. The relationship between self-reported and performance-based measures of cognitive empathic processes is unclear. Design Cross-sectional analysis of the association between medical students’ empathy scale scores and their empathic performance in a visuospatial perspective-taking (VPT) task. Participants Undergraduate medical students across two European medical schools (n=194). Primary and secondary outcome measures Two self-report empathy and one performance-based perspective-taking outcome: Jefferson Scale of Physician Empathy (JSPE); Empathy Quotient (EQ); Samson’s level-1 VPT task. Results Higher scores on the ‘standing in patient’s shoes’ subscale of the JSPE were associated with a lower congruency effect (as well as lower egocentric and altercentric biases) in the VPT (B=−0.007, 95% CI=−0.013 to 0.002, p<0.05), which reflects an association with better capacity to manage conflicting self-other perspectives, also known as self-other distinction. Lower egocentric bias was also associated with higher scores on the ‘social skills’ EQ subscale (B=−10.17, 95% CI=−17.98 to 2.36, p<0.05). Additionally, selection of a ‘technique-oriented’ clinical specialty preference was associated with a higher self-perspective advantage in the VPT, reflecting greater attentional priority given to the self-perspective. Conclusions We show that self-assessment scores are associated with selected performance-based indices of perspective taking, providing a more fine-grained analysis of the cognitive domain of empathy assessed in medical student empathy scales. This analysis allows us to generate new critical hypotheses about the reasons why only certain self-report empathy measures (or their subscales) are associated with physicians’ observed empathic ability.