Premature ventricular contraction (PVC) is the most common form of arrhythmia that can be life-threatening at any time. Fast and accurate use of computer-aided detection of PVC is critical for both doctors and patients. In this paper, we propose a new method for PVC detection based on abnormal eigenvalues and decision tree. We choose composite areas, amplitudes and intervals as feature parameters to identify heartbeat types. The method was tested in the publicly available MIT-BIH arrhythmia database with an accuracy of 99.6%, a sensitivity of 97.5%, and a specificity of 99.7%. The effectiveness of the proposed method is proved by comparison with other methods.