The multi-attribute health status (MAHS) approach to developing indirect utility measures assumes that health preferences are formed from the simultaneous consideration of multiple health dimensions. We propose an alternative view that theorizes the formation of health preferences is mediated by global impressions of quality of life (QL). This study compared the theory of global health preference formation (GHPF) with the MAHS approach for explaining time trade-off utilities. A total of 1432 cancer patients completed the EORTC QLQ-C30 and valued their own current health. A mediation analysis was performed using latent variable models relating health preferences to QLQ-C30 domains. Founded on the MAHS approach, Model I described health preferences using the physical, role, cognitive, emotional, and social functioning domains (PF, RF, CF, EF, and SF, respectively) and the fatigue (FA), pain (PA), and nausea/vomiting (NV) symptom domains. Model II related the QL domain to the same functioning and symptom domains. Consistent with the GHPF framework, Model III purported that QL mediated associations of health preferences with functioning and symptoms. Ignoring QL, health preferences were related to PF (b=0.041/p=0.050), SF (b=0.057/p=0.028), and EF (b=-0.054/p=0.001). Model II: QL was positively related to RF (b=0.222/p=0.001), EF (b=0.116/p=0.001), SF (b=0.257/p<0.001), FA (b=0.245/p<0.001), and PA (b=0.123/p=0.001). Model III: QL was positively related to health preferences (b=0.117/p<0.001). Controlling for QL, the only functioning or symptom domain related to health preferences was EF (b=-0.067/p<0.001). Significant indirect effects representing differences between direct effect estimates for Models I and III were observed for RF (b=0.026/p=0.001), EF (b=0.014/p=0.002), SF (b=0.030/p<0.001), FA (b=0.029/p=0.003), and PA (b=0.014/p=0.002). Model III provided a significantly better fit than Model I (p<0.001). The MAHS approach yields misspecified models of health preferences since QL mediates associations of the latter with functioning and symptoms. Our framework has far-reaching implications for utility assessment and warrants further research.