Background: Normal values are commonly used in medical settings and generally determined based on the empirical distribution of normal subjects. Methods for longitudinal measures are not well developed. We evaluated alternative methods to estimate quantiles for longitudinal quality of life (QOL) measures following prostate cancer treatment. Methods: QOL was measured both pre-treatment and up to two years post-treatment in 2046 men. Quantile regression, Baysian quantile regression and linear regression with z-score quantile estimation were considered, as well as methods to adjust for repeated measures. K-fold cross-validation was used to evaluate model calibration. Post-treatment QOL was modeled using pre-treatment QOL, treatment, time since treatment and clinical factors as predictors. Results: Quantile regression methods proved more flexible and generally had better calibration in the case of skewness, floor and ceiling effects, and differential effects by quantile. However, as it independently estimates each quantile function, it can create inconsistencies which linear techniques avoid. Weighting by inverse number of observations per person slightly improved calibration.