In this paper, we present a method for clustering spatiotemporal pass trajectories in soccer games by using multiscale structural matching and rough clustering techniques. The problems of irregular sampling intervals and irregular sqeuence lengths are dealt by multiscale matching, and the problem of possible local disturbance of a dissimilarity matrix caused by matching failure is dealt by rough clustering. Experimental results demonstrate that the combination of these methods leads to successful discovery of interesting pass patterns in the real soccer game records, for example, side-attack after complex pass transactions, and zigzag pass transactions, as well as the ability of the proposed method as a grouping method for general spatio-temporal data.