Studying individuals with extreme phenotypes could facilitate the understanding of disease modification by genetic or environmental factors. Our aim was to identify Huntington's disease (HD) patients with extreme symbol digit modality test (SDMT) scores. We first examined in HD the contribution of cognitive measures of the Unified Huntington's Disease Rating Scale (UHDRS) in predicting clinical endpoints. The language‐independent SDMT was used to identify patients performing very well or very poorly relative to their CAG and age cohort. We used data from REGISTRY and COHORT observational study participants (5,603 HD participants with CAG repeats above 39 with 13,868 visits) and of 1,006 healthy volunteers (with 2,241 visits), included to identify natural aging and education effects on cognitive measures. Separate Cox proportional hazards models with CAG, age at study entry, education, sex, UHDRS total motor score and cognitive (SDMT, verbal fluency, Stroop tests) scores as covariates were used to predict clinical endpoints. Quantile regression for longitudinal language‐independent SDMT data was used for boundary (2.5% and 97.5% quantiles) estimation and extreme score analyses stratified by age, education, and CAG repeat length. Ten percent of HD participants had an extreme SDMT phenotype for at least one visit. In contrast, only about 3% of participants were consistent SDMT extremes at two or more visits. The thresholds for the one‐visit and two‐visit extremes can be used to classify existing and new individuals. The identification of these phenotype extremes can be useful in the search for disease modifiers.