Abstract: To solve the outlier detection problems where outliers highly intermix with normal data, a general variance-based outlier detection model (VODM) is presented, in which the information of data is decomposed into normal and abnormal components according to their variances. With minimal loss of normal information in the model, outliers are viewed as the top k samples holding maximal abnormal information in a dataset. The VODM is a theoretical framework, and then, the principal curve is introduced as an algorithm of it. Experiments carried out on abnormal returns detection in stock market show that the VODM is feasible.