We develop a life cycle model that features food consumption, exercise, and deviation from reference BMI, which represents local social norms, to rationalize spatial concentration and the educational gradient in body mass in the US. BMI is determined by caloric balance and affects health and medical spending, the probability of survival, and the level of utility. We find that occupational strenuousness, food prices, medical expenditures, and labor income explain only a modest share of BMI differences across regions and education groups. Preference differences matter, but they account for the patterns only when amplified by social norms. To demonstrate the policy relevance of the social norm, we introduce a GLP-1 treatment policy targeting individuals with the highest food preferences. We find substantial direct effects on average BMI, with reductions of up to more than 6 pounds. Importantly, we document meaningful spillover effects arising from endogenous adjustment of reference BMI that further amplify these effects. The BMI reductions translate into substantial life expectancy gains for both treated individuals and untreated individuals through spillover effects.