OBJECTIVES: Patient-reported outcomes (PROs) are widely used in clinical research. However, results on score differences may be hard to interpret if clinicians are unfamiliar with the assessment tools and lack benchmarks for interpretation of results. This study aims to estimate clinical and social benchmarks for interpretation of score differences on the SF-36 Health Survey and to test whether the interpretation depends on score level and patient background characteristics. METHODS: Using survival and logistic regression models, we reanalyzed data from three US cohort studies: the Medical Outcomes Study (N=3,445), the Medicare Health Outcomes Survey (N=78,183), and the QualityMetric 2009 Norming Study (N=4,040). Outcome variables were mortality, hospitalization, current inability to work, and loss of ability to work. RESULTS: Benchmarks were robust across types of disease and for people with or without chronic disease. However, the benchmarks did vary according to age and score level. A one-point lower score on the PF, GH and PCS scales was associated with a 1.05-1.09 relative risk (RR) of mortality for the typical chronic disease patient, with stronger associations in the younger age groups. For several scales (PF, RP, BP, GH, VT, SF, and RE), the associations with mortality also depended on score level, with stronger associations in the lower score ranges (i.e. patients in worse health). A one-point lower score on the PF, RP, BP, GH, VT, SF, and PCS scales implied a 1.02-1.04 RR of hospitalization, a 1.07-1.15 RR of being unable to work, and a 1.04-1.06 RR of losing the ability to work. CONCLUSIONS: Statistical associations with outcomes such as mortality, hospitalization, and work ability can be used as benchmarks to interpret score differences on the SF-36.