Abstract: Using nuclear quadrupole resonance (NQR) to detect hidden explosive has many advantages, such as high accuracy, low false alarm rate, non-magnetic pollution, and so on. However, NQR signal is very weak, obliterated in the noise, and difficult to recognize, and the practical use of NQR is restricted by this. According to the characteristics of NQR signal, a signal process method was presented and realized based on wavelet transform. It shows that this method can eliminate the interferences of external noise, and improve the detection accuracy of the system comparing with other methods.