A technique of PVC detection is described in this paper. The procedures of PVC analysis include noise detection, QRS detection, feature extraction and QRS wave classification. The interference is detected in hierarchy ways. A new non-linear transformation of ECG waveform is given to be taken as the QRS detection function. Taking this QRS detection function, the influence of noise and interference can be reduced greatly. By the technique of QRS template matching, QRS waveforms are sorted into different clusters in QRS wave form space. Then a linear classifier is used to identify the medical meaning of these clustes. Before the linear classifier is applied, the QRS waveforms are classified according to some special decision rules. 38 half-hour standard database is used to test the technique presented in this paper. The result shows t hat accuracy rate of PVC detection is better than 95%.