A conceptual runoff model is studied in this work. Especially the residuals of the model (the differences between the computed hydrograph and the recorded one) are examinated. The density function and the autocorrelation function of the residuals are estimated and tested.

The model must be calibrated for each new catchment, to which it is applied. Therefore a criterion of the goodness of fit between the computed and the recorded hydrograph is required. Some criteria of fit have been examined concerning their sensitivity to changes in the model parameter setting.

Conclusions of the work are: The residuals of the model are neither stationary nor independent and normally distributed. A classification based on the different physical processes, which govern the discharge, makes the residuals of each class more stationary and in some sense more normally distributed than the residuals of the material without any classification. Furthermore a criterion of goodness of fit, based on the classification above resembles the subjectively judged fit more than the simple sum of squares criterion, which has become practice in applications of runoff models.