Abstract: Hydrological variation is occurring in many areas, which often causes the ‘pollution’ of hydrological data and even makes the traditional hydrological frequency calculation in a difficult position. Although a few approaches to this issue have been proposed, some problems, such as different solutions obtained by different methods, masking effect, impact of autocorrelation, and overlapped variation information, are still barriers to developing hydrological design theories in the changing environment. In this paper, a TFPW-BS-Pettitt method is presented for the issue, which combines trend-free pre-whiting (TFPW), binary segmentation (BS) and Pettitt test to deal with the autocorrelation and masking effects occurring in change-point detection of streamflow data series. And hydrological frequency analysis is used to verify its accuracy in detection of change-point locations by comparing the data distributions before and after the change-points. Results indicate that this method can alleviate the autocorrelation effect on change-point detection through TFPW processing and eliminate the masking problem by BS algorithm, which makes multiple change-points detected effectively. Disadvantage in this new model is that the influence of data length on Pettitt test still remains. Especially when the length of a data sequence beyond the marginalized change-point is less than 5, this point is easily missed and thus a treatment of the edge effect should be done before detection. In addition, it is also found that when only change-points in mean exist in the sequence, an equal near-natural streamflow series can be obtained through calculation according with the data series containing the changes. And it may be used in the design to alleviate the impact of data lack in practical frequency analysis. When change-points in mean and variance both exist, such a series can be obtained only for the piece of data before the first change-point in the variance.