Abstract We developed decision trees (J48 algorithm) to predict the distribution of the pike (Esox lucius). Based on a historical data, 75 sampling sites were considered for the pike in 7 stream basins in Flanders (Belgium). In total, 108 instances were available in the given sites. The measured variables consisted of a combination of the structural-habitat, physical–chemical and biological variables (biomass, abundance and presence/absence of the pike). The predictive power of decision trees was assessed on the basis of the number of Correctly Classified Instances (CCI %) and Kappa statistic (k). In order to reduce the noise in the data and improve the predictive results with regard to complexity and accuracy of the predictions, different Pruning Confidence Factors (PCFs) were tested. The obtained results showed that the prediction of the pike (based on presence/absence data) was acceptable in terms of two model evaluations (CCI>70% and k>0.40). The habitat variables had more contribution to the prediction of distribution of pike relative to the water quality ones. The developed model presented a logical relationship between distribution of the pike and distance from the source, slope and followed by depth. These models can as such become essential tools to encourage river managers to make the necessary investments and/or activity changes as desired by society.