Assessment of hydrological model results is an ordinary task being continuously conducted by every modeller now and then. Depending on the modelling ideology expert judgement or statistical criteria are given greater or less weight. Usually trusts to statistical criteria are much higher when the model parameters are estimated using methods of automated calibration basically by streamflow observations. In other cases when a modeller’s mind is open to accept his model’s failures due to model imperfection in processes representation and there is an aim to improve it the expert judgement becomes invaluable.

In this presentation several cases of inconsistency between the statistical and expert judgement criteria would be described by the examples of the Hydrograph model application in different environments. Some examples would reveal how the streamflow may be mimicked completely failing to describe other watershed processes with the statistical criteria glorifying the model, or how the deep immersion to the detailed watershed data may change the expert judgement of model results.

The experience shows that at the stage of studying the processes and implementing a model at new environment the expert judgement is the most important; when the model is set up for some specific conditions the statistical criteria may be used.