In a recent article in this journal, Smith offers additional evidence to support his claim that the test-retest reliability of willingness to pay measures increases along with willingness to pay because people take more time to consider their answers for the more highly valued (and therefore more 'expensive') goods. Unfortunately, by repeating a common misconception about what reliability actually measures, he overlooks an alternative explanation for the relationship he observed; namely, that subject variation increases with willingness to pay and that it is this, rather than any reduction in measurement error, that explains his findings. We show that 75% of the increase in reliability comes from increases in subject variation (that is different views about the value of good health), and that the relationship between measurement error and willingness to pay is not as simple as Smith suggests. However, our critique of Smith's paper should not be construed as criticism of the ideas being explored. We need to better understand the responses people give to contingent valuation exercises. Such understanding has to be based on a better appreciation of what reliability is and on more robust testing of alternative hypotheses.