we calculated the seismometer's self-noise level with 1000 consecutive hours of data, and proposed using the probability and statistics method to analyze the calculation results. the analysis results are demonstrated by three expressions:multi-hour result overlap, probability statistics on single frequency point and probability statistics in subband. Through the comparison and analysis, we found that it is easy to track anomalies but hard to get the non-uniform distribution characteristic for the first expression. For the second one, it is easy to get the non-uniform distribution characteristic of the self-noise level at the fixed frequency point without obvious difference in orders of magnitude. the third method is to obtain a high probability value corresponding to the center frequency point and having a magnitude difference. In the case of a small amount of abnormal data, the abnormal results are not reflected in the latter two ways.