Obesity encapsulates the increased risk of disease and premature death associated with excess fat, but it is usually measured using a simple function of weight and height known as the Body Mass Index (BMI). Economists use the BMI to determine the prevalence of obesity within the population, to estimate the “disease burden” created by obesity, to establish the additional medical spending attributable to obesity, and to measure the current and potential effects of government policy on obesity. Estimates will be biased if the measurement error from using BMI to proxy for obesity is correlated with other covariates. In this paper we use a flexible function of percent body fat and several metabolic factors to construct an obesity index to predict death and disease. We use this index to show that using BMI leads researchers to misattribute obesity-related health outcomes to other factors such as aging, family health history, alcohol consumption, and household income. Moreover, we show how these biases necessarily generate the opposite implications for studies on the causes of obesity. Specifically, using BMI leads researchers to underestimate the effects of aging, family history of diabetes, alcohol consumption, and household income on obesity. We find no significant evidence of measurement error bias associated with smoking or insomnia. We also show that, compared with our obesity index, BMI generates substantially more false negative predictions of diabetes.