Sampling matching is a new methodology for the selection of study samples from pools of opt-in respondents. This methodology addresses the primary substantive and technical issues of how large, but unrepresentative, panels can be used to construct representative study samples for particular target populations. The procedure uses a listing or enumeration of the population that can be obtained from large scale consumer and voter databases that have been developed in recent years. The existence of such data has not been exploited in previous Internet research. On both a theoretical and a practical level, this approach substantially improves upon existing weighting procedures. As validation, we show how this procedure performed in predicting the outcome of the 2005 California special election.