Due to randomness in measured modal frequencies, the present study develops the probabilistic detection method for abnormal change of bridge structures’ modal frequencies based on the theories of probability and statistics. The BPNN-based correlation models between modal frequencies and structural temperature are established, so as to eliminate the temperature effects in modal frequencies. Then the cumulative distribution functions of the modal frequencies after temperature effect eliminating are estimated using Kernel density estimation method. The estimated cumulative distribution functions are converted to Q statistics based on the inverse function of standard normal distribution. At last the mean value control charts of Q statistics are constructed to detect the abnormal change of modal frequencies. The hypothesis testing results show that Q statistics follow Gaussian distributions. Accordingly, the non-normality of modal frequencies with temperature effect elimination is handled to meet the requirement of random variables’ normality in control charts. The results reveal that the probabilistic detection method of control chart have good sensitivity to the abnormal data of bridge’s modal frequencies. The proposed method can provide a reference for the analysis and application of health monitoring data of long-span bridges.