Sequential pattern mining is the one most concentrated and applied in sequence mining research, it gives a frequency based view of the correlations between elements contained in the sequences. However, when we consider domain knowledge within the data mining process, the frequency based criterion becomes less interesting since most of the frequent sequences might have already been confirmed, and the most interesting sequences might not be the sequences corresponding to existing knowledge, but be the sequences contradicting existing knowledge that reflect unexpected behaviors. In this paper we introduce the problem of finding unexpected behaviors within the context of sequence mining. We first give formal descriptions of belief base and unexpected sequences, we then introduce unexpected sequential patterns and unexpectedness rules that depict unexpected behaviors within the sequences. We also propose the USER approach for mining unexpected sequential patterns and rules from a sequence database with respect to a given belief base. Our experimental results show that both of the quantity and the quality of the unexpected sequences extracted by the USER approach are improved in comparison with the frequent sequences extracted by general sequential pattern mining approaches.