In the present study, classification trees (CTs-J48 algorithm) were used to study the occurrence of roach in rivers in Flanders (Belgium). The presence/absence of roach was modelled based on a set of river characteristics. The predictive performance of the CTs models was assessed based on the percentage of Correctly Classified Instances (CCI) and Cohen's kappa statistics. To find the best model performance, a 3- fold cross validation techniques was applied on the dataset. The effect of Pruning Confidence Factors (PCFs) was examined on the reliability and model complexity. Based on the obtained results, the induced model could predict well the presence/absence of roach in the rivers. The highest overall means of two model performances showed that the models were reliable. When analyzing the ecological relevance of CTs, it seemed that the structural-habitat variables were more the main predictors than the water quality ones to predict the occurrence of roach in rivers. In particular, the distance from the source and width contributed more to the prediction of roach while among water quality variables, only electric conductivity was relatively important in this regard.