We describe the results of a survey designed to assess the test-retest reliability of a method of establishing willingness to pay. Willingness to pay values for a hypothetical intervention were elicited from a randomly selected, population sample by face-to-face interview on three occasions over a period of 5 weeks. Test-retest reliability was assessed by intraclass correlation and by generalizability analysis. Reliability was acceptable but not substantial, and there was a statistically significant shift in mean value between first and second assessments. The greatest source of variation in values was the participants. There was also a substantial interaction between time and participants, suggesting that some respondents changed their answers at follow-up. The results were sensitive to the high valuations provided by four of the participants, however.