Cost-utility analyses conducted in Phase III clinical trials have obtained preference values for health outcomes typically from one of two sources - direct interviews of trial subjects, or a generic health status index to impute preference weights to subjects. But the interview approach has been criticized over concerns about administrative feasibility and generalizability. The generic instrument approach usually requires the investigator to make subjective judgments in mapping a subject’s clinical profile into a (pre-computed) preference weight. This paper introduces a third approach and tests its viability in a trial comparing rheumatoid arthritis therapies. Specifically, we: (a) identified a “surrogate sample” of arthritis patients (N=123) who were clinically comparable to trial subjects; (b) elicited their preference (utility) ratings for their own current (arthritis-affected) state of health; (c) estimated a multivariable regression model relating the surrogate-sample utility scores to these patients functional status, clinical symptoms, and demographics; (d) used these regressions to assign utility scores to trial subjects at each scheduled encounter; and (e) computed encounter-specific mean utility scores for each treatment group. The conclusions are: (1) it is feasible to use a surrogate sample to estimate regression models that satisfactorily relate patient utility to readily collectible predictor variables; (2) these models can be applied to assign plausible utility scores to subjects in a Phase III clinical trial; (3) resulting ranking of therapies is in accord with clinical findings from present trial; and (4) this strategy may be especially attractive for cost-utility analyses that pool data from multiple trials.