According to the irregularity and complexity of roller bearing fault, and vibration signals can reflect the occurrence and development of the fault, a roller bearing fault diagnosis method based on wavelet packet transform (WPT) and Sample Entropy (SampEn) is proposed. SampEn is a measure that quantifies the complexity of a signal and has the advantage of being less dependent on time series length. The original bearing vibration signal is decomposed by wavelet packet transform. The sample entropy of the resultant wavelet packet coefficients are served as feature vector. In the classification, the support vector machine method is used to identify the different faults. Experiments are conducted on roller bearing with three different fault categories and several levels of fault severity. The experimental results indicate that the proposed approach could reliably identify the different fault categories. Thus, the proposed approach has possibility for bearing fault diagnosis.