The fuel-use decisions of households in developing economies, because they directly influence the level of indoor air quality that these households enjoy (with its attendant health effects), provide a natural arena for empirically assessing latent preferences towards the environment and how these evolve with increases in income. Such an assessment is critical for a better understanding of the likely effects of aggregate economic growth on the environment. Using household data from Pakistan we estimate Engel curves for traditional (dirty) and modern (clean) fuels. Our results provide empirical support for the household choice framework developed in Pfaff, Chaudhuri and Nye (2002a), which suggests that even if environmental quality is a normal good, non-monotonic environmental Engel curves can arise. Under plausible assumptions about the emissions implied by fuel use, our estimates yield an inverted-U relationship between indoor air pollution and income, mirroring the environmental Kuznets curves that have been documented using aggregate data. We then demonstrate, through a simple voting model, that this household-choice framework can generate aggregate EKCs even in a multi-agent setting with heterogeneous households and purely external environmental effects.