Background Despite its monogenic cause, the cognitive problems in Huntington’s Disease are heterogeneous. This is particularly pronounced in the premanifest (preHD) stage of the disease. The existing tools cannot explain all of the heterogeneity observed nor can they accurately inform the progression of these aspects of the disease at an individual level. Method We have developed a data driven mathematical model using the cross-sectional data from the first ENROLL-HD data cut to classify preHD participants into two groups. Machine learning methods were used to identify variables of interest that were inputted into a support vector machine which was used to classify participants. The model was validated in a local subset of the PREDICT-HD cohort. Results The support vector machine used performance on four cognitive tests to classify participants into two distinct groups: group one who perform worse on cognitive measures and group two who perform better. Importantly, the two groups did not differ on measure of age, CAG repeat number and disease burden measures. We also found a significant genetic association between membership of the cognitively better group and the minor alleles of NCOR1 and ADORA2B through supplementary genotyping analysis. Conclusion This work suggests that participants can be classified into distinct groups on the basis of their cognitive performance alone, even in the premanifest stage of HD. There may be genetic interactions within the HD gene pathway that explain some of the cognitive heterogeneity that we see in preHD patients.