Abstract: A multi-stage detection method of narrowband signal is given, which combines the double thresholds detection, frequency tracking, and expert system together. The factors that influence the detection performance of narrowband signals are analyzed. The analysis shows that this method can automatic detect narrowband signal with high detection probability and low false probability at low SNR without knowing their amplitude, position, and number of signals in advance. The proposed method is also robust in dealing with sea trial data and it is fit for real time processing because of small computation amount.