Three satellite rainfall products (3B42V7, IMERGV05/V04/V03, and China hourly Merged Precipitation Analysis product) are evaluated using measurements from a dense rain gauge network as reference in Guangdong Province, China from April 2014 to December 2016. The three products are compared with gauges in annual, monthly, daily and hourly accumulation, and at gridded, sub‐regional and regional scales. An error decomposition approach is employed to separate the total bias into Hit, Miss and False components. Overall, the CMPA estimate is the best in agreement with gauge observations. The improvement of IMERGV05 over 3B42V7 is notable, especially in reducing the hit bias and missed precipitation, resulting in better detection of the light rain and heavy rain. The enhancement of IMERG is more significant at the 3‐hr scale than daily scale. IMERGV05 shows better performance than its previous versions especially for IMERGV03, which had an abnormal blocky pattern mainly from July to September over the Pearl River Delta area. The three products have different error characteristics and show large spatial variations. 3B42V7 and IMERG have large areas of overestimation in the mountainous areas and underestimation in the coastal areas, while CMPA is characterized by an alternate distribution of small positive and negative values. The overestimation of precipitation is partially attributed to the positive hit biases that falsely estimate the precipitation from moderate intensity to heavy rainfall by the three products.