In this paper, we introduce the class of k-partite episodes, which are time-series patterns of the form ?A1, . . . ,Ak? for sets Ai (1 = i = k) of events meaning that, in an input event sequence, every event of Ai is followed by every event of Ai+1 for every 1 = i < k. Then, we present a backtracking algorithm Kpar and its modification Kpar2 that find all of the frequent k-partite episodes from an input event sequence without duplication. By theoretical analysis, we show that these two algorithms run in polynomial delay and polynomial space in total input size.