Random-digit-dialing surveys in the United States such as the Behavioral Risk Factor Surveillance System (BRFSS) typically poststratify on age, gender and race/ethnicity using control totals from an appropriate source such as the 2000 Census, the Current Population Survey, or the American Community Survey. Using logistic regression and interaction detection software we identified key "main effect" socio-demographic variables and important two-factor interactions associated with several health risk factor outcomes measured in the BRFSS, one of the largest annual RDD surveys in the United States. A procedure was developed to construct control totals, which were consistent with estimates of age, gender, and race/ethnicity obtained from a commercial source and distributions of other demographic variables from the Current Population Survey. Raking was used to incorporate main effects and two-factor interaction margins into the weighting of the BRFSS survey data. The resulting risk factor estimates were then compared with those based on the current BRFSS weighting methodology and mean squared error estimates were developed. The research demonstrates that by identifying socio-demographic variables associated with key outcome variables and including these variables in the weighting methodology, nonresponse bias can be substantially reduced.