Abstract: To improve performance of endpoint detection in noisy environment is an important issue in automatic speech recognition(ASR),especially in actual noisy environments.The performance of conventional endpoint detection methods based on short-time energy and zero-crossing rate is unsatisfactory in environments of low SNR.Approximate entropy is a new statistical method of complexity measurement in measuring time series complexity.It is quite stable with changing data length and has strong anti-interference ability.In this paper,a method of noisy speech endpoint detection based on approximate entropy(ApEn)is proposed.Simulation results indicate that the method has good performance even in low SNR circumstances.