Fine grid resolution and temporal scales are crucial to analyze the hydrologic impacts from extreme floods for a watershed. A regional scale atmospheric model such as MM5 is useful to achieve this purpose because ground observation stations are rarely located within watersheds. However, the achieved data set by a model should be validated against observation data before it is used. In this study, historical global reanalysis atmospheric data (NCAR/NCEP) was dynamically downscaled to a 3km resolution over the Upper Feather, Yuba, and American River watersheds using the Fifth-Generation NCAR / Penn State Mesoscale Model (MM5). To validate the reconstructed precipitation, historical data from California Data Exchange Center (CDEC) stations were first used. However, the data that can be obtained from them is coarse throughout the watersheds so historical data from PRISM (parameter-elevation regressions on independent slopes model) were also used. The hourly increment precipitation data found at the CDEC stations in each watershed are used to verify the temporal accuracy of the model simulation, and the PRISM simulations are used to verify the monthly basin-average precipitation and its spatial distribution over each watershed.