Individual differences in vulnerability to sleep loss can be considerable, and thus, recent efforts have focused on developing individualized models for predicting the effects of sleep loss on performance. Individualized models constructed using a Bayesian formulation, which combines an individual's available performance data with a priori performance predictions from a group-average model, typically need at least 40 h of individual data before showing significant improvement over the group-average model predictions. Here, we improve upon the basic Bayesian formulation for developing individualized models by observing that individuals may be classified into three sleep-loss phenotypes: resilient, average, and vulnerable. For each phenotype, we developed a phenotype-specific group-average model and used these models to identify each individual's phenotype. We then used the phenotype-specific models within the Bayesian formulation to make individualized predictions. Results on psychomotor vigilance test data from 48 individuals indicated that, on average, ∼85% of individual phenotypes were accurately identified within 30 h of wakefulness. The percentage improvement of the proposed approach in 10-h-ahead predictions was 16% for resilient subjects and 6% for vulnerable subjects. The trade-off for these improvements was a slight decrease in prediction accuracy for average subjects.