OBJECTIVES: Numerous cross-sectional studies reported that high body mass index (BMI) is associated with poorer health-related quality of life (HRQL), while only few studies also provide evidence for impaired HRQL in underweight individuals. Our objective was to investigate the nonlinear relationship between BMI and HRQL in an elderly general population sample using semiparametric regression methods. METHODS: We analyzed data from 4562 individuals aged 65 years or older, living in the region of Augsburg, Southern Germany. The data come from the KORA-Age study which is based on a postal follow-up of individuals who participated in the population-based MONICA/KORA surveys S1-S4. Health-related quality of life was measured using the German EQ-5D index. For our multivariable regression analyses, we used a generalized additive model (GAM) to estimate the functional form of the relationship between BMI and HRQL after adjusting for confounding factors. RESULTS: We found a significant nonlinear (inverse U-shaped) relationship between BMI and HRQL after adjusting for gender, age, sociodemographic factors and comorbid conditions. The maximum HRQL was observed at a BMI of about 25 kg/m2. Our estimates indicate that going from a BMI of 25 to 35 is associated with a EQ-5D utility loss of about 5.2 units (corresponding to 0.28 of standard deviation). On the other hand, underweight individuals with a BMI of 18 had an average impairment of 5.8 units compared to a BMI of 25. Subgroup analyses showed that the inverse U-shaped relationship is more pronounced in individuals ≥ 75 years than in individuals aged 65-74 years. In particular, the effect of low BMI on HRQL is more important in the older age group. CONCLUSIONS: Generalized additive models are an adequate method to estimate the nonlinear relationship between BMI and HRQL in the elderly population. With increasing age, low BMI has similar impairments in HRQL as overweight.