I use a novel empirical approach to study heterogeneities in labor market flows in the U.S. Using data on short-term labor force history, I develop Bayesian logit models that capture large variations in the job finding probabilities not only of the unemployed but of the non-participants, as well. A decomposition of these variations by time periods suggest that recent employment has a bigger impact on current job finding probabilities than contemporaneous search behavior. Furthermore, I document that among prime age men, non-participants are almost as likely to start working as active job-seekers. The gap narrows even further in recessions, for the job finding probability of the unemployed-looking tends to fall disproportionately.