A challenge in hydrological studies in the Canadian Prairie region is to find good-quality meteorological data because many basins are located in remote regions where few stations are available, and existing stations typically have short records and often contain a high number of missing data. The recently released North American Regional Reanalysis (NARR) data set appears to have potential for hydrological studies in data-scarce central Canada. The main objectives of this study are: (1) to evaluate and utilize NARR data for hydrologic modelling and statistical downscaling, (2) to develop methods for estimating missing precipitation data using NARR data, and (3) to investigate and correct NARR precipitation bias in the Canadian Prairie region. Prior to applying NARR for hydrological modelling, the NARR surface data were evaluated by comparison with observed meteorological data over the Canadian Prairie region. The comparison results indicated that NARR is a suitable alternative to observed surface meteorological data and thus useful for hydrological modelling. After evaluation of NARR surface climate data, the SLURP model was set up with input data from NARR and calibrated for several watersheds. The results indicated that the hydrological model can be reasonably calibrated using NARR data as input. The relatively good agreement between precipitation from NARR and observed station data suggests that NARR information may be used in the estimation of missing precipitation records at weather stations. Several traditional methods for estimating missing data were compared with three NARR-based estimation methods. The results show that NARR-based methods significantly improved the estimation of precipitation compared to the traditional methods. The existence of NARR bias is a critical issue that must be addressed prior to the use of the data. Using observed weather station data, a statistical interpolation technique (also known as Optimum Interpolation) was employed to correct gridded NARR precipitation for bias. The results suggest that the method significantly reduces NARR bias over the selected study area.