The authors describe the use of Internet-based surveys to predict the results of 72 races in the US elections of November 2000. The authors describe their results, the methods they used, including the use of both demographic and "propensity score" weighting to correct for substantial biases in the raw, unweighted data, and the investments which were needed to achieve these results. They also caution readers not to assume that identical methods can be used with equal success in other countries or to measure all other variables, and stress the need for continuing research to improve online research methods in the future.