Four methods of statistical downscaling of daily precipitation were evaluated on three catchments located in southern, eastern, and central China. The evaluation focused on seasonal variation of statistical properties of precipitation and indices describing the precipitation regime, e.g., maximum length of dry spell and maximum 5-day precipitation, as well as interannual and intra-annual variations of precipitation. The predictors used in this study were mean sea level pressure, geopotential heights at 1000, 850, 700, and 500 hPa, and specific humidity as well as horizontal winds at 850, 700, and 500 hPa levels from the NCEP/NCAR reanalysis with 2.5° × 2.5° resolution for 1961–2000. The predictand was daily precipitation from 13 stations. Two analogue methods, one using principal components analysis (PCA) and the other Teweles-Wobus scores (TWS), a multiregression technique with a weather generator producing precipitation (SDSM) and a fuzzy-rule-based weather-pattern-classification method (MOFRBC), were used. Temporal and spatial properties of the predictors were carefully evaluated to derive the optimum setting for each method, and MOFRBC and SDSM were implemented in two modes, with and without humidity as predictor. The results showed that (1) precipitation was most successfully downscaled in the southern and eastern catchments located close to the coast, (2) winter properties were generally better downscaled, (3) MOFRBC and SDSM performed overall better than the analogue methods, (4) the modeled interannual variation in precipitation was improved when humidity was added to the predictor set, and (5), the annual precipitation cycle was well captured with all methods.