Abstract: In order to apply the satellite and gauges rainfall datasets in hydrological simulation and forecasting, the TRMM PR 3B42 V6 product is evaluated by ground-based observation rainfall in alpine and gorge region of Jin-sha River Basin, which involves 3-hourly, daily and monthly temporal scales. Concretely, the linear regression analysis is used to analyze correlation of the two datasets. The empirical orthogonal function and singular value decomposition methods (EOF-SVD) are adapted to examining the spatial patterns of precipitation. The indices of relative deviation (B_ias), false alarm ratio (R_FA) and the probability of detection (P_D) are applied to evaluating the accuracy of the satellite dataset. The results indicate that a significant linear correlation is derived between the two datasets at different spatiotemporal scales, but the correlation coefficient will be decreased with the reduction on temporal scales. Good consistency is achieved in the overall spatial patterns of precipitation, even though local discrepancies are found in high altitude and high- gradient areas. The TRMM 3B42 data are accurate about 36.08% within 10% of the station locations. The altitude shows certain influence on the indices of R_FA and P_D, which demonstrate increasing on R_FA but decreasing on P_D with elevation rising. Overall, the accuracy of TRMM 3B42 data declined with elevation rising; the periods rainfall error decreased with temporal scale increasing and the light rain have the greatest contribution to satellite deviation but the heavy rainfall is the second. These results suggest that merge and correction would be necessary when the satellite data are used for hydrological simulation and forecasting on finer temporal scales in similar investigation areas.