A heteroscedastic double-hurdle model is used to investigate household butter consumption in the United States. Results suggest that failure to incorporate heteroscedastic errors may lead to unreliable elasticity estimates. Decomposition of the effects of variables leads to insightful information and makes the double-hurdle model a more useful tool in micro demand analysis. Larger and higher-income households are more likely to consume butter than others and also consume more, but income elasticity is very small. Age, region, and seasonality are among the other significant determinants of household butter consumption.