Hydrological forecasts are essential, both for the prevention of flood damages and for water resources planning. In Northern Sweden, snowmelt plays an important role in the formation of runoff. Spring flood forecasts have been carried out since the middle of the 1970s, using the HBV runoff model. In the HBV model, the snow pack is simulated from interpolated daily observations of point precipitation and temperature. The acquirement of representative data is often difficult as the highest precipitation occurs at high altitudes, which are sparsely populated and difficult to reach. Remote sensing data on the snow pack should thus be important as an additional source of information. The project presented in this report had two aims:- To modify the HBV model to include remote sensing data as input to the simulations.- To evaluate the influence of such data on the accuracy of simulated runoff.The remote sensing data available to the project came from NOAA-AVHRR images, which provided data on snow covered area under cloud free conditions. The evaluation was carried out for a medium-sized catchment in the mountainous region in the northwest of Sweden. Satellite data were available for five different years. To facilitate the use of remote sensing data, a gridded version of the HBV model was developed. Procedures and criteria were developed to automatically calibrate the HBV model against both runoff and snow cover data. This was done to minimise the risk of compensating errors in the parameter values of the model.Due to clouds, remote sensing data are not available on a regular basis. Consequently they were not utilised as model input in the same sense as precipitation and temperature. When available, they were instead used to correct errors in the simulated snow pack. Model routines were developed to compare observed and simulated snow cover and to automatically make the corrections.For the evaluation, the data set was divided into two periods. The model was calibrated independently for each period and verified for the other. The results were contradictory and not conclusive. For the first period, the precipitation appeared to be systematically overestimated, which led to compensating errors in the parameter fitting and an erroneously modelled snow distribution. Attempts to correct the snow pack for the second period thus failed. For the second period, there were no apparent systematic errors in the precipitation input. After calibrating the model for this period, satellite data could be used to considerably improve the accuracy in the runoff simulations for the first period. The overestimation of precipitation and thereby the snow pack could be corrected for.The most effective way to overcome the problem of systematic errors in the input data for the calibration period is longer data records. Another possibility is more sophisticated calibration routines than the ones developed within this project. A grid by grid comparison of modelled and observed snow cover showed systematic deviations. It indicates that there are improvements to be made in the snow model, and that remote sensing data can be useful in such work