Résumé : We present an integer programming model for human leukocyte antigen (HLA) association studies based on the parsimony criterion. The model is simple, compact, easy to implement, and able to handle datasets containing up to 200 phenotypes. Computational experiments carried out on patients affected by psoriasis and severe alopecia areata show that the model is consistent with the experimental haplotype frequencies, showing, for the considered diseases at least, a high reliability of the predictions. These promising results provide perspective on computer-aided association studies and encourage the development of efficient exact computational approaches for haplotype estimation. © 2010 American Society for Histocompatibility and Immunogenetics.