Objective ECG analysis diagnosis system mainly consists of two phases: waveform recognition and intelligent diagnosis. Accuracy and reliability of ECG waveform recognition determine the diagnosis and treatment of heart disease. A waveform recognition algorithm based on geometric characteristics is proposed in this paper to improve the detection rate and accuracy of waveform recognition. Methods Firstly, signals are preprocessed to improve noise ratio by digital filtering algorithm. Secondly, we calculate certain geometric characteristics such as the slope and the movement tendency of points with improved second derivative. Finally, we can automatically recognize the onset, the offset, the peak and the trough of T wave, P wave, QRS complex with the actual physical characteristics of ECG waveform on the basis of the previous results. Results Statistical analysis shows that our method can recognize the ECG wave fast and effectively. We also compare this method with other ECG waveform recognition algorithms. The results indicate that the proposed method results in better performance in the sensitivity (Se) and positive predictive (PP) than other methods. Conclusions The proposed method makes improvement on current ECG waveform recognition methods.