To identify accurately the location of rock fractures events, it is necessary to detect the arrival time of acoustic emission signals recorded by the relevant AE machine. However, the arrival times of signals containing high-amplitude noise are difficult to be detected accurately. In this study, an improved STA/LTA method based on denoising technique and Multi Scale (MS) theory is proposed, whereby we apply first wavelet denoising and MS theory to overcome the limitations of noise and then the STA/LTA is applied to pick arrivals in different scales. The performance of improved method is also validated by the synthetic signals with four SNRs. The results show the picking accuracies of arrival time based on the improved STA/LTA method are credible even through at very low SNR conditions.