Abstract: The Split Spectrum Processing (SSP) technique has proved its ability in reduction of interference noise in ultrasonic nondestructive testing for coarse grained materials. However, the results of SSP algorithms are not sufficient stable since they are sensitive to the filter bank and filter parameters. And it is not clear how the technique use the filter outputs available. In this paper, the author has theoretically discussed the statistical phase response characteristic of filter outputs for ultrasonic testing. Thus, a new SSP algorithm based on phase standard deviation is created. The performance has been examined for both computer simulated and experimental data, and compared to commonly used minimum algorithm. The phase standard deviation algorithm has proved its superior effect and is less sensitive on the number of filters.