We disentangle and study the relative importance of different risk preferences in explaining extended warranty purchases and the high premia paid for them. Empirical and behavioral research on insurance is at odds with whether diminishing returns (curvature of the utility function), or loss aversion and nonlinear probability weighting lead to observed consumer behavior. This lack of consensus is primarily due to the inability of standard choice data to separate different risk preferences, and the consequent need to rely on strong parametric assumptions. We design two conjoint studies (consistent with simultaneous and sequential decision making) with choices about washing machines and extended warranties, where subjects are given failure probabilities and repair costs. Using stated choice data from the surveys, we can nonparametrically identify product and risk preferences. We find that loss aversion is significantly more important than curvature and probability weights in explaining extended warranty choices. Importantly, failure to decompose risk-averse behavior into that arising from curvature, loss aversion, and probability weighting leads to lower washer prices and profits. These findings are robust to alternate reference point assumptions. We rationalize the premia paid for warranties by exploring retailer incentives to price discriminate, and test the theory on complementary goods pricing. Finally, based on counterfactual analysis, forcing separate retailers to sell washers and extended warranties is not necessarily welfare enhancing as cited in the media and previous literature.