Abstract: A discrete hidden Markov model (HMM) for ECG (electrocardiogram) signal feature extraction is built, which solves problems of ECG signal feature extraction in body sensor network and considers the feature partition of ECG waveform. Based on the proposed HMM, methods of expert annotation selection, lead selection, normalization of observation data, triple initial value selection, and training data quantity selection are customized. Finally, the HMM model parameters are trained by using the Baum-Welch algorithm, and the ECG signal feature is extracted by using the Viterbi algorithm. Simulation results show that this feature extraction method for ECG signal based on HMM has lower complexity, higher accuracy, and better timeliness, which is suitable for processing nonlinear and dynamic changing ECG signal on-line and can satisfy the performance requirements of ECG signal feature extraction in body sensor network.