An expert system for supervised classifier design has been presented. A number of heuristics have been imple- mented to shrink the search in the large space of candidate designs. The CommonKADS methodology has been used to specify the knowledge and reasoning requirements of the system. Alzheimer’s diagnosis has been used as a case study and optimal results were found. Different standard datasets for classification have also been used to evaluate the performance of the system. In all cases, the expert sys- tem provided solutions with an error rate very close to the best reported results by other researchers on the field. Fur- thermore, the heuristics used to guide the search have been proven to be effective since in most cases it was not neces- sary to try an excessive number of design proposals before finding a suitable one.