Distributed hydrological models have been applied in various watershed hydrological processes. They are often combined with numerical weather and climate prediction, which make them need enormous calculation. In recent years, the progress of GPU technology makes the ordinary computer to perform efficient and inexpensive parallel computing. This paper presented the GPU implementation of data interpolation, runoff generation and unit hydrograph calculation in parallel compution. A recursive non-parallel implementation of Muskingum river -routing method was also presented. Based on the common notebook computer with NVIDIA GPU/CUDA and C# language, the parallel simulation of rainfall-runoff process in the Yihe River Basin by the Xinanjiang Model based distributed hydrological model indicates that the performance of parallel execution of precipitation spatial interpolation and Xinanjiang discharge calculation has a speed of 8~9 times of that from a common CPU based C # execution. The recursive Muskingum method was also 0.5~0.9 times faster than the traditional calculation using a routing order table.