As an important task of data mining

outlier detection has attracted much attention.We discuss the issues of outlier definition and detection based on rough set theory.We propose a new definition for outlier-rough sequence outlier

and the corresponding outlier detection algorithm RSOD.The algorithm constructs three kinds of sequences exploiting the notions of knowledge entropy and significance of attribute in rough sets

and detects outliers by analyzing changes of the elements in the sequences.We compare algorithm RSOD with the current outlier detection algorithms on UCI data sets.And experimental results show that our method is effective for outlier detection.