Most current quantitative precipitation nowcasting (QPN) methodologies are deterministic in nature and convey little realistic information regarding the uncertainty in the forecast. For flood and streamflow forecasting applications it is important to understand and accurately represent this uncertainty. This will allow a greater deal of confidence in the use of precipitation forecasts for hydrological purposes. This paper presents results obtained using a stochastic Bayesian precipitation nowcasting scheme. The nowcast methodology is hierarchical in nature and, as such, allows the use of other observed meteorological parameters (e.g. Doppler winds) as a constraint on the stochastic process. This produces a physically based statistical nowcast. The scheme produces a range of nowcasting solutions using a series of initial fields of radar reflectivity, resulting in a distribution of short-period forecasts. This paper examines the nature and interpretation of the distribution of QPNs. We present results of a study of the variability of point nowcasts (meteograms) generated and areal accumulations. Areal precipitation forecasts are used as input to a simple theoretical lumped hydrological catchment model to investigate the impact of the nowcast distribution on streamflow forecasts. This investigation provides information on how forecast uncertainty can be propagated through such models, and how one may best handle distributions of QPNs of this nature in a hydrological context.