To test the accuracy of downscaled precipitation estimates from numerical weather prediction models, we have carried out a set of experiments to study what space and time scales are appropriate to obtain downscaled precipitation forecasts with maximum skill. For operational streamflow forecasting in the Upper Colorado Basin, six-hourly mean areal precipitation (MAP) and temperature values (MAT) are used to run the NWSRFS (National Weather Service River Forecasting System) models. In our research, we have found that downscaled temperature fields have significant skill even up to 5 days of forecast lead-time, and there is practically no skill in the downscaled precipitation forecasts. In this paper we present a set of numerical experiments to understand the effects of space and temporal aggregations that affect the skill of downscaled precipitation forecasts. The 6-hrly MAP values used in the NWSRFS are a set of calibrated values (weighted mean) derived from daily and hourly stations distributed throughout the Basin. The questions we seek to answer are, is it more useful to downscale first to the individual stations and then aggregate to derive the MAPs or directly downscale to the MAPs? How does the skill vary when we downscale precipitation to 6-hrly values, to daily values, to 3-day averages and 5-day averages both at the point scale (stations) and regional scale (for example, MAPs). Initial results show that, higher forecast skills (measured using the RPSS, ranked probability skill score) are obtained for the 6-hrly time scale for the regional scale than downscaling to individual stations. This result is intuitive, because averaging over space helps to filter the noise present in the high frequency precipitation signals. However, if we move to a daily time scale, the skill obtained at individual stations is similar to those obtained for example at the MAP scale (regional). We are also carrying out further experiments to look into the sensitivity of spatial averaging on the skill of downscaled precipitation.