In order to investigate how cognitive skills develop in the course of expertise in Japanese fencing, regular (expert) and substitute (junior expert) players of a men's university varsity team were compared on performances on (I) a paper-pencil test of rules and concepts (Test a),(II) convergent problem solving tasks such as to predict a scorer's winning trick from a video just before it occurs (Test c-2), and (III) divergent problem solving tasks such as to judge players' skill from their postures (Test b-1), and to detect defects in them (Test b-2). Unexperienced college students also participated in the experiment in part. Both the experts and junior experts knew the rules and concepts of Japanese fencing well, and their performances were much better than the estimated baseline. Their performances in convergent problem solving were also equally well. In divergent problem solving, however, the experts were better than either the junior experts or the unexperienced. These results suggest that divergent problem solving skills need a longer time to develop.