Abstract: Mining frequent closed episodes from an event sequence is an important task. The existing research work is based on the support definition of minimal occurrences and the breadth-first search strategy, which unavoidably leads to the issues such as over-counting the occurrences of an episode and generating a huge number of candidate episodes. In this paper, a novel algorithm FCEMiner is proposed to mine frequent closed episodes from an event sequence, which employs the support definition of both minimal and non-overlapping occurrences and the depth-first search strategy. Moreover, FCEMiner utilizes the non-closed unanimity of special forward extension to skip redundant closure checking and narrow down the search space of frequent closed episodes. Both theoretical study and experimental evaluation confirm that FCEMiner is able to effectively discover frequent closed episodes from an event sequence.