This paper introduces a weighting approach based on a propensity score calculation in order to reduce the bias inherent in online or Internet surveys. By running parallel survey treatments, one online, framed as the experiment, and one via random digit dial telephone methodologies, framed as the control experiment, it is possible to eliminate observed differences between online and telephone surveys and many of the unobserved differences that randomization controls for. We will present both the theoretical justification for the methodology and results of a multitude of experiments that we have run using parallel surveys. Finally we will present results showing that a well constructed propensity model can be ported from survey to survey if there is no parallel RDD telephone survey. This methodology allows one to project survey results from online surveys to the general population.