Abstract: About three billion people meet their household energy needs with wood and other solid fuels. In recent years epidemiologists have come to view indoor air pollution from poorly ventilated cooking fires as a major cause of illness and death, particularly among women and children. To date, social scientists have paid scant attention to the behavioral antecedents of indoor air pollution. In short, we do not understand why people use household energy technologies that kill them. This dissertation uses household survey data from Peru to explore three questions related to this puzzle. First, I use a reduced form demand framework to study factors affecting fuel choice. This analysis leads to three salient conclusions. I find that the cross-section data approach that has dominated empirical analysis of household fuel choice systematically overstates the role of income in determining fuel choice, probably due to omitted variable problems. A second finding of the demand analysis is that fuel choice is highly responsive to fuel price; in particular, I find that increases in fuel price are sufficient to explain the drop in fossil fuel use in Peru during the 1998-2002 period of the survey panel. Finally I present tentative evidence suggesting that endogeneity in fuel prices and---to a lesser extent---household income may also be a problem in much of the existing demand literature, and I develop methods for consistent estimation in the face of such endogeneity. The second part of the dissertation looks for evidence of social learning in fuel choice patterns. I begin by showing observables such as income and infrastructure cannot explain the spatial distribution of fuel use, which suggests that other forces may be at play. I develop a simple learning model embedded in the random utility maximization (RUM) framework. Parameter estimates are more consistent with a learning model than with a non-learning peer effects model. The final set of questions pertains to the health impacts of indoor air pollution. I show that the regression approach generally used in the epidemiological literature gives biased estimates of the health consequences of indoor air pollution. Empirical results based on a health production function approach suggest that this bias is downward (towards zero) in direction and substantial in magnitude.