OBJECTIVES: The statistical analysis of health utilities and health-related quality of life (HRQL) scores poses various challenges due to the distributional properties such data commonly exhibit. These include skewness and heteroscedasticity caused by the bounded scale. Various analytical approaches have recently been proposed for use in cross-sectional studies, however, less attention is paid to longitudinal designs. We examined the use of beta regression models to analyze HRQL data over time using two empirical examples. METHODS: The HRQL measures employed in our empirical examples were the generic SF-6D and the disease-specific Stroke Impact Scale (SIS). Data came from the German KORA cohort study and from a clinical setting, respectively. We fitted mixed and marginal beta models and explain the conceptual difference between these two model classes, namely the population-averaged and the subject-specific perspective. We compared overall fit and predictive accuracy of the models to the commonly used linear mixed model (LMM). RESULTS: The SF-6D data were highly skewed to the left and the SIS data exhibited a pronounced ceiling effect. In both examples, the beta distribution fitted the data better than the normal distribution. Beta regression accounted for the fact that predicted values must fall into the bounded support of the scales and overall fit measures suggested that the mixed beta model was superior to the LMM (AIC: -2723 vs. -2441 and -904.1 vs. -376.4). However, mixed beta regression underestimated the mean at the upper part of the distribution. Adjusted group mean scores from the marginal beta model were nearly identical to those derived from the LMM. CONCLUSIONS: Longitudinal beta regression models are potential candidates to analyze HRQL over time since they account for the specific characteristics such data typically have. However, our results show that in practice, estimates may only differ slightly from those of commonly used methods.