Arsenic contamination of drinking water is a serious public health issue in many areas of South and South East Asia. One study estimates that in Cambodia over 100,000 people are exposed, with the majority of those living in Kandal Province. In this thesis we present 3 original empirical studies focused on estimating nonmarket values for reduced arsenic risk water, based on primary data collected in May 2013. We also present a review paper which discusses the various economic techniques which have typically been used to estimate welfare values for cost-benefit analysis of mitigation strategies or appraisal of drinking water standards. The first empirical paper presents the results of a discrete choice experiment (DCE)to estimate willingness to pay (WTP) values for reduced arsenic water. We discuss the results of scale-extended latent class choice models and underlying differences in preferences and choice consistency. We find that a reduction in the permissible limit on arsenic in drinking water may best represent underlying household preferences for risk. The second empirical paper presents the results of a split sample choice experiment focusing on differences between money (WTP) and labour contributions (WTWork) as payment vehicles in terms of choice behaviour and attribute non-attendance. We find that the results from the two experiments are relatively consistent which reinforces our results from the previous chapter that focuses on WTP measures alone and adds credibility to the large numbers of DCEs conducted in rural areas of developing countries. The final empirical paper examines actual household behaviour relative to an arsenic testing and education campaign run by a local NGO. We find that the vast majority of households change their drinking water source upon being informed that it is unsafe. On average households that switch increase their expenditures. In doing so however they also reduce the amount of time spent collecting water which limits the use of expenditure changes as an approximation of welfare values.