Abstract: With the aid of the observation data obtained from 33 precipitation stations during 1989—2013 in the western region of Zhejiang Province within the Taihu Lake watershed, a predictive model for the annual precipitation spatial estimation has been developed based on the generalized additive models (GAM), by use of them an annual precipitation data set with 1 000 m×1 000 m spatial resolution in the study area was gained. By several statistical indexes making analysis of influences of different predictive variables on the precipitation spatial estimation results, the accuracy and reliability of GAM based on the interpolation method were evaluated, especially comparing the differences between two kinds of the models in estimating the annual precipitation space whether or not it selects an elevation as the predictive variable. The analysed and estimated results show that, after taking the elevation as the predicative variable, whether the GAM performance or the precipitation spatial estimation accuracy were improved apparently, which can reasonably reflect the precipitation spatial distribution influenced by the topographic factors in the study area. In addition, it is found from the GAM partial effect graphs that the annual precipitation amount obviously decreased with the increase of the latitude and remarkably increased with the increase of the altitude in the study area, however, there were no obvious changes in the annual precipitation amount with the increase of the longitude.