Rainfall index insurance is a theoretically attractive financial product that has achieved only limited adoption. This paper seeks to understand the structure of demand for rainfall index insurance in India. We develop two approaches to estimating households’ valuation of rainfall insurance and evaluate them against an experiment in which fixed prices are randomly assigned. The first approach uses a simple structural model of index insurance demand that includes basis risk–the possibility that policy-holders may suffer a negative shock yet receive little or no payout. We use survey data from members of an insurance pilot in Gujarat, India to fit the model and estimate the willingness to pay (WTP) for rainfall insurance coverage. Relative to the choices we observe at randomly assigned fixed prices, the structural model significantly overestimates demand. Our second approach uses a Becker-Degroot-Marschak (BDM) methodology to empirically elicit WTP from potential insurance customers at the time of marketing. We find that BDM does a better job of predicting fixed price purchasing behavior, but the distribution of stated willingness to pay has large mass points at focal points. Finally, we directly compare the two approaches and find the theoretical model has weak predictive power for WTP as elicited by BDM. We explore which household characteristics are correlated with WTP and determine that recent experiences with rainfall and insurance are important factors not captured in our static model, suggesting that learning dynamics may be a promising direction for future analyses.