RISE is an algorithm that combines rule induction and instance-based learning (IBL). It has been empirically verified to achieve higher accuracy than state-of-the-art representatives of its parent approaches in a large number of benchmark problems. This paper investigates the con-ditions under which RISE’s bias will be more appropriate than that of the pure approaches, through experiments in carefully controlled artificial domains. RISE’s advantage compared to pure rule induction increases with increasing concept specificity. RISE’s advantage compared to pure IBL is greater when the relevance of features is context-dependent (i.e., when some of the features used to describe examples are relevant only given other features’ values). The paper also reports lesion studies and other empirical observations showing that RISE’s good performance is indeed due to its combination of rule induction and IBL, and not to the presence of either component alone.