Introduction:The different ways for selecting single nucleotide polymorphisms have been related to paradoxical conclusions about their usefulness in predicting chronic fatigue syndrome even when using the same dataset. Objective: To evaluate the efficacy in predicting this syndrome by using polymorphisms selected by a supervised approach that is claimed to be a method that helps identifying their optimal profile. Materials and methods: We eliminated those polymorphisms that did not meet the Hardy-Weinberg equilibrium. Next, the profile of polymorphisms was obtained through the supervised approach and three aspects were evaluated: comparison of prediction accuracy with the accuracy of a profile that was based on linkage disequilibrium, assessment of the efficacy in determining a higher risk stratum, and estimating the algorithm influence on accuracy. Results: A valid profile (p<0.01) was obtained with a higher accuracy than the one based on linkage disequilibrium, 72.8 vs. 62.2% (p<0.01). This profile included two known polymorphisms associated with chronic fatigue syndrome, the NR3C1_11159943 major allele and the 5HTT_7911132 minor allele. Muscular pain or sinus nasal symptoms in the stratum with the profile predicted V with a higher accuracy than those symptoms in the entire dataset, 87.1 vs. 70.4% (p<0.01) and 92.5 vs. 71.8% (p<0.01) respectively. The profile led to similar accuracies with different algorithms. Conclusions: The supervised approach made it possible to discover a reliable profile of polymorphisms associated with this syndrome. Using this profile, accuracy for this dataset was the highest reported and it increased when the profile was combined with clinical data.