Because the distributed fiber fence system based on the phasesensitive optical timedomain reflectometer is very sensitive to its ambient environments, high false alarm rate is a quite normal phenomenon which is very annoying and difficult to avoid. In this paper, different wavelet decomposition methods are compared for extracting the singularity information buried in the noisy intrusion signals. A comprehensive judgment method is proposed which combines several different wavelet decomposition methods to reduce the system′s false alarm rate. The experiment results show that different wavelet decomposition methods can extract different detail information at different levels, thus it can provide much more sufficient information and conditions for the decisionmaking process. The false alarm rate of the new method can be lowered to none or once per 24 hours, which is much lower than the other methods based on a single wavelet analysis way. It can efficiently improve the system′s robustness and its decision′s confidence level.