Discrete choice experiments (DCEs) are increasingly used for ex-ante evaluations of environmental policies but their validity and reliability are largely untested in low-income settings. My thesis examines whether DCEs provide valid and reliable estimates of welfare impacts in these contexts and comprises a systematic review and three field tests of DCE in a new REDD+ (Reducing Emissions from Deforestation and forest Degradation) project and national park in eastern Madagascar. I first conducted a systematic review of empirical evidence on the reliability and validity of DCEs when valuing non-market environmental goods from the past 13 years. I found limited and mixed evidence and mostly recommend the use of more qualitative, interdisciplinary and deliberative approaches. The review informed the design of my empirical DCE survey, which elicited local people’s preferences for the right to clear forests for swidden agriculture, cash payments, and support for improved rice farming. I used this survey and rigorous qualitative debriefings to investigate three aspects of reliability and validity of DCEs. I first surveyed households who have varying experience of restrictions on swidden agriculture to explore the validity of DCE in estimating the costs of conservation restrictions ex-ante. I found that experience of forest protection matters; households who have been exposed to forest protection for a comparatively longer period had significantly higher welfare costs for restricting forest clearance than those who are less experienced. I conclude that although DCE can elicit current preferences in my study context, DCE is not a valid ex-ante tool for estimating compensations for such a long-term and complex intervention. I then used a within-subject design to evaluate whether giving respondents more time to deliberate influences DCE responses. I found that deliberation significantly affected individual-level preference parameters and marginal willingness-to-accept (WTA) estimates, but the effect depended on their literacy. Illiterate respondents’ WTA significantly increased post-deliberation while literate respondents’ decreased. Although respondents’ decisions to revise their choices mostly stemmed from more reflection, I also found evidence of strategic behaviour among more literate respondents. Greater time to deliberate may improve welfare estimates with illiterate respondents, but may have risks when used with others. Finally, I compared responses to the WTA and willingness-to-pay (WTP) DCE formats. While it is well recognised that the choice of format can influence welfare estimates, I showed that it also influenced which attributes are significant. The WTA format is more appropriate in my study context based on three criteria; its content validity, its acceptability to respondents, and respondents’ ability to pay. I also found that local people were generally averse to state protection and strongly demand secure forest tenure. My thesis makes an important methodological contribution to the advancement of DCE techniques. It also has major implications for how forest conservation policy may be devised in low-income countries, including devolution of secure forestland tenure to local people and genuinely negotiating conservation with forest users.