Based on the gauges data of 2423 meteorological stations in China, continuous statistical metrics (correlation coefficient: R, root mean squared error: RMSE, mean absolute error: MAE, relative error: RE) and classified statistical metrics (probability of detection: POD, false alarm ratio: FAR, bias score: Bias, equitable threat score: ETS) are used to analyze the accuracy of Global Prediction Measurement (GPM) precipitation products from three dimensions including different spatial scales, different time scales and different precipitation intensity so as to explore the applicability of GPM satellite precipitation products over mainland China. The main results show that from different spatial scales, GPM precipitation has high observation accuracy in all regions, and the correlation coefficient (R) values of 72% stations exceede 0.7, with the best in East China and relatively poor in Northwest China. RE is concentrated in 0-20%. The accuracy of different altitude-zone elevations show that the overestimation of GPM is more obvious in low altitude (< 2000 m) and high altitude (> 4000 m) regions, and the applicability of GPM data is relatively good in mid-altitude regions (2000-4000 m). In terms of different temporal scales, the total annual precipitation of GPM is consistent with that of rain gauges, and the R is 0.75, but there is a certain deviation in the amount of precipitation, with RMSE being 6.15 mm·d-1. The consistency between GPM precipitation products and rain gauges is better from January to October, with the R above 0.7, slightly lower in November and December. The error value in summer is higher than that in winter, and the RE is positive mostly. In addition, the accuracy results of different precipitation intensity suggests that POD decreases with the increase〖JP2〗 of precipitation intensity. GPM precipitation products have a better detection ability for “moderate rain” intensity, while the detection ability for “light rain” and “heavy rain” is slightly weaker. 〖JP〗