This paper provides a comprehensive and systematic framework for evaluating a wide range of statistical information in order to optimize the design of dual frame surveys. Formulae for the total mean square error of dual frame survey estimators are derived for general stratified multistage sample designs under a model which incorporates nonsampling bias and variance terms. A general procedure for se letting resource allocations which minimize the total survey error for a fixed budget is developed. The methodology is applied to the Current Population Survey for which a number of uses of the procedure are illustrated.