When a sequence variation is found in a candidate gene for a disease, it is important to establish whether this change is neutral or responsible for the observed disorders in a patient. To answer this question, in the absence of further experimental investigations, several simulation programs have been proposed to predict whether a nonsynonymous single-nucleotide polymorphism is likely to have or not have a deleterious effect on the phenotype. In this work, we tested two such programs, PolyPhen and SIFT, using two kinds of targets. The first ones concerned the products of the hemoglobin and glucose-6-phosphate dehydrogenase genes, which are abundantly documented. The second concerned two systems for which much less information is available: (a) the TNFRSF1A gene, implicated in tumor necrosis factor receptor-associated periodic syndrome, and (b) the MEFV gene, which is believed to be involved in familial Mediterranean fever. Our data suggest that, from a practical point of view, these programs should not be used to decide, in the absence of other tests or arguments, whether the sequence variation found in a patient is or is not responsible for the disease. The consequence of an erroneous prediction may be disastrous in the perspective of genetic counseling.