OBJECTIVES:  To estimate the logical inconsistency rate from the scores given to EQ-5D health states by a sample of  the Thai general population. METHODS: A total of 12 health sets (11 health states for each set)  were used with one set allocated to each respondent. Respondents were requested to  value 11 states using the Ranking and Visual Analog Scale (VAS) methods and 10 states using the Time Trade-off (TTO) method.  Respondents were requested to rank 11 health states. Then, they were asked to rate the health states on a scale from 0 (worst imaginable) and 100 (best imaginable). Finally, respondents were asked to compare being in  a particular state for a period of years and being in a perfect health for a shorter time (TTO). Logically inconsistent values are defined as cases where the respondent gave a higher value to the logically poorer state. The numbers of inconsistencies in each interview method was re-expressed as an inconsistency rate.  A Poisson regression was used to analyse these rates. RESULTS: A total of 1409 respondents; 644 male and 765 female were interviewed by 43 trained interviewers during May-August 2007. Respondents’ mean age was 44.6 years (min. 20 yrs. and max. 94 yrs.). Age and educational level have a significant effect on the inconsistency rate  (P-value=0.001), the rate of inconsistency increases by 1% for each additional year of life and in the respondents with lower educational level. Gender, number of children, duration of the overall interview process and VAS score for own health have no significant effects on the inconsistency rate. CONCLUSIONS: The inclusion of the scores from older respondents with lower education should proceed with caution. The inconsistency rate can be used to guide the exclusion of the respondents with inconsistent responses to achieve the best models estimating the scores for health states.