Abstract: When the Mann-Kendall (MK) method is used to detect the trend in serially correlated hydrological series, it is often difficult to obtain an accurate result, due to the influences by significant higher-order autocorrelation components or the trend damage in removing the lag-1 autocorrelation in the series. Aiming at this difficulty, this paper presents a hybrid MK test model combining a modified over-whitening (MOW) process and assembling the change point tests in regression and variance and the detrended methods to segment the original series and provide a set of suitable sub-series to over-whitening. In this model, the over-whitening is calculated in sections to remove the high-order autocorrelation while destroying the original trend component as little as possible. The test results of the runoff series at Linjiacun, Shenmu, Zhaoshiyao and Hengshan hydrologic stations indicate that the segmental over-whitening process can retain the trend change in the original series when eliminating its significant lag-high autocorrelation and ensure more accurate results in the MK trend test on over-whitened series.