Health care expenditures have been increasing rapidly. Economic evaluation can be used to aid decision making on resource allocations to secure a more efficient use of scarce resources. In cost-utility analysis, one method used to measure health outcomes is the Quality adjusted life year (QALY). Given the wide differences in clinical settings, health systems and religious beliefs, "utility" scores should be derived from the local population. This thesis aims to estimate population-based preference scores for health from the Thai general population. The generic health description EQ-SO is used as a proxy to describe health. This measure was selected because it has been translated officially into Thai and the measure seems to be straightforward to use. A representative sample was randomly recruited using a stratified four-stage sampling method. A series of pilot studies were conducted to develop the interview protocol based on the Measurement and Valuation in Health (MVH) protocol. A group of interviewers were employed and extensively trained to interview the respondents. A sample of 1,409 Thai respondents was interviewed during May - August 2007 in 17 provinces in face-to-face interviews. Eighty-six health states, classified into twelve sets, were used in the interview. logical inconsistency was identified when a higher score was given to a poorer state. The greatest number of inconsistent responses was identified in the scores derived using the Time trade-off (nO) interview. A Negative binomial regression model was used to analyse the determinants of the numbers of inconsistencies. Elderly respondents and those with a lower education level tend to make more inconsistent responses. A Random effects model was used to estimate the model to predict the preference scores. The best model was chosen on the basis of logical inconsistency in the predicted scores, model robustness, parsimony and the responsiveness of the predicted scores. The best model is the model using the variables from Dolan 1997 model estimated from the scores given by the respondents with fewer than 11 inconsistencies. The model still suffers from heteroskedasticity, and floor and ceiling effects were identified. The Thai scores and the scores derived from respondents in the other five countries were extensively compared to examine the extent of the differences. It seems that the Thai scores are more similar to those of the UK. A costutility analysis of the prevention and control measures for cervical cancer in Thailand was used to demonstrate the difference of cost per QALYs if the scores from other countries were used to approximate the Thai preferences. The thesis makes a number of contributions. The modelled scores are the first original population-based preference scores on health derived from the Thai general population. The determinants of logical inconsistency were examined, as well as an exploratory qualitative interview to learn the strategies that respondents employed to cope with the preference interview. Three reasons are identified to explain the high level of inconsistent responses. Respondents may: (1) have difficulties imagining themselves living in the hypothetical states; (2) use only part of the given information in the health cards or add other information to assist their decisions; and (3) have difficulties in trying to understand the elicitation methods, especially the no. Including the inconsistent responses had, to some extent, significant impacts on the model specifications and the modelled scores. Exclusion of the scores from the highly inconsistent respondents was justified because the scores may not represent their preferences towards health. The results from this thesis should be taken into account for future surveys to be successfully administered. Close collaborations with the field coordinators and arrangement of appropriate interview settings contribute greatly to the success of the survey.