No hydrologic model is perfect. Differences between what is simulated by a hydrologic model and what actually occurs may be important to account for if valid probabilistic hydrologic forecasts are to be produced using imperfect hydrologic models. There are many causes of these differences including measurement errors, model errors and forecast errors. The effect of these is to cause biases not only in the mean value of the forecasts but also in the shape of the forecast probability distribution. Operational procedures have been developed for the National Weather Service River Forecst System (NWSRFS) to account for the effects of these differences on streamflow forecasts. These procedures are being tested in the upper Monongahela basin demonstration project. The techniques and recent experience with them will be presented