The differential diagnosis of endemic hemorrhagic fevers in tropical countries is by no means an easy task for medical practitioners. Several diseases often overlap with others in terms of signs and symptoms, thus making this diagnosis a difficult, error-prone process. Machine Learning algorithms possess some useful qualities to tackle this kind of pattern recognition problems. In this paper, a neural-network-based approach to the differential diagnosis of Dengue Fever, Leptospirosis and Malaria, using the Adaptive Resonance Theory Map (ARTMAP) family is discussed. The use of an Artificial Immune System (CLONALG) led to the identification of a subset of symptoms that enhanced the performance of the classifiers considered. Training, validation and testing phases were conducted using a dataset consisting of medical charts from patients treated in the last 10 years at Napole´on Franco Pareja Children Hospital in Cartagena, Colombia. Results obtained on the test set are promising, and support the feasibility of this approach.