A new sampling approach for case-control studies offers a flexible alternative to frequency matching. In the "randomized recruitment" method, subjects are individually randomized to be recruited or not on the basis of investigator-imposed recruitment probabilities that can depend on both disease status and values of covariates already available or ascertained in a screening interview. When there is prior information about the odds ratios associated with the screening variables, such a design can achieve "probability matching," without the well-known disadvantages that encumber traditional matching. The method can also be used to enlarge the relative size of subsamples of interest. Following randomized recruitment, a modified logistic regression analysis allows unbiased estimation of effects associated with all variables studied, including the "matching" variables. One can also readily fit an additive model. The method is illustrated by developing the recruitment probabilities required for probability matching on age, sex, and cigarette smoking status in an ongoing study of lung cancer and exposure to radon progeny in which smoking cases are undersampled.