OBJECTIVES: Several empiric algorithms have been developed to derive preference estimates from the SF-36. The objective of this study was to compare the validity and responsiveness of SF-36 derived preference estimates to directly elicited preferences in persons with persistent asthma. METHODS: We used data from a one-year clinical trial of adult asthmatics to derive preferences from the SF-36. Preferences were estimated using five published algorithms for converting SF-36 scores to non-choice based preference estimates. Derived preferences were compared to directly elicited visual analog scale (VAS) values. RESULTS: The average VAS value at baseline was 0.65 (SD=0.19), 0.83 (SD=0.14) at the end of the study and the average change score from baseline to end of study for the VAS was 0.17 (SD=0.20). The differences between the VAS value and the derived preferences ranged from -0.07 to 0.02 at baseline. Three of the five 95% confidence intervals (CIs) for the difference between derived preferences and VAS values at baseline included zero. At the end of the study the mean of the VAS was higher than the means for all of the derived preference methods. The difference between the averages for the VAS and the derived preferences ranged from 0.01 to 0.11 and only one of the 95% CIs for the difference included zero. The change scores for the VAS preferences were greater than the derived preferences (Differences from 0.07 to 0.13). None of the 95% CIs for the difference in change scores between VAS preferences and derived preferences crosses zero. CONCLUSIONS: The derivation methods produce valid and responsive measures of patient preference. However, the derived preference values differ from each other and directly elicited preference values. Differences in the distributions of the directly elicited and derived preferences will affect inferences and can lead to differing conclusions in a cost-utility analysis.