Similarity of two time series is a good measure of the reliability of the sensor. In the conventional method, the coherence statistic of two auto-spectral densities is applied. This paper is concerned chiefly the improvement of reliability testing method using coherence statistic. A Monte Carlo simulation method has been presented for making optimal decision statistic. The reliability testing using multi-sensor checking method also has been developed. Two experiments were performed in artificial and real data set. The improvement of capability for reliability checking was verified in first experiment using artificial data. The validity was shown in second experiment using real sensor measurements in a dam.