In the search for drugs against schizophrenia and depression without extrapyramidal side effects, compounds that selectively antagonize the dopamine D3 receptor subtype are thought to be a solution. In order to create a model with which the D3 activity can be predicted and that can generate new ideas for future synthesis, we performed a comparative molecular field analysis (CoMFA). In our model 30 ligands were described quantitatively in the GRID program, and the model was optimized by selecting only the most informative variables in the GOLPE program. We found the predictive ability of the model to increase significantly when the number of variables was reduced from 25110 to 784. A Q2 of 0.65 was obtained with the final model, confirming the predictive ability of the model. By studying the PLS coefficients in informative 3D contour plots, ideas for the synthesis of new compounds can be generated.