Building on sibling difference regressions, a new IV strategy for estimating heterogenous interaction effects in obesity is proposed. Identification depends on a testable restriction: standardized direct responses to variations in own covariates differ across different types of siblings. The sibling difference model is a testable special case. Using the NLSY79 children dataset, the null hypothesis of the sibling difference model of no sibling interaction effect or contextual effect on weight or BMI is not rejected using a specification test which is simple to implement. However, when we estimate the more general model we find small but statistically significant peer effects. Brothers are estimated to have a small negative reaction to an increase in their siblings' obesity whereas sisters have a small positive reaction.