Theoretical noise trader models suggest that uninformed traders can impact market prices. However, these models' conclusions depend crucially on the assumed specification for noise trader demand. This research seeks to empirically determine the appropriate demand specification for uninformed traders. Using commercial market sentiment indices as proxies for noise trader demand, Granger causality models are estimated to examine the linear linkages between sentiment and futures returns. The models strongly suggest that noise traders are positive feedback traders (i.e., extrapolative expectations) with relatively long memories.