We describe a new method of overcoming problems inherent in peculiar velocity surveys by using data compression as a filter with which to separate large-scale, linear flows from small-scale noise that biases the results systematically. We demonstrate the effectiveness of our method using realistic catalogs of galaxy velocities drawn from N--body simulations. Our tests show that a likelihood analysis of simulated catalogs that uses all of the information contained in the peculiar velocities results in a bias in the estimation of the power spectrum shape parameter $\Gamma$ and amplitude $\beta$, and that our method of analysis effectively removes this bias. We expect that this new method will cause peculiar velocity surveys to re--emerge as a useful tool to determine cosmological parameters.