BACKGROUND Expert panel data are often used to collect resource use, effectiveness, and quality of life data when alternative methods are impractical due to cost, timing, or for ethical reasons. OBJECTIVE: To compare several methods of deriving EQ-5D valuations from expert panel data. METHODS: A panel of 7 physicians provided EQ-5D valuations for 5 health states associated with atrial fibrillation: (State 1) Well in atrial fibrillation; (State 2) Minor disability in normal sinus rhythm; (State 3) Minor disability in atrial fibrillation; (State 4) Major disability in atrial fibrillation; (State 5) Major disability in normal sinus rhythm. Three methods of combination were used to estimate utilities: a) averaging calculated utility scores b) the modal dimension valuation c) the modal overall valuation. The resulting utility weights for each method were compared across health states, and on internal consistency (expected ordering of the health states as above). RESULTS: The valuations from the methods were similar within states for State 1 (0.85 to 0.86), State 3 (0.49 to 0.52) and State 5 (-0.07 to 0.09), but disparate utilities resulted for State 2 (0.52 to 0.66) and State 4 (-.17 to 0.52). The averaging and modal dimension methods gave the expected ordering. The modal dimension method, however, estimated a very large step (ie. 0.68) in utility between states of major disability. The averaging method gave results that were the most linear between states. CONCLUSIONS: Lack of a gold standard methodology makes synthesizing expert panel utility valuations problematic. In this instance, averaging gave results consistent with expected ordering and with no unexpected large drops in utility.