This paper contributes to the literature about estimating asset correlation in two ways. First, we compare the performance of different estimation approaches in a simulation study. By doing so, we provide knowledge about the behavior of the applied estimators, which is an important precondition for the interpretation and robustness of the estimation results. Second, we present a novel data set from which to estimate asset correlation: the loss data of residential mortgage-backed security (RMBS) deals. Our data set is largely made up of the most toxic RMBS deals that sparked the subprime crisis. Contrary to the widely held view, our analysis reveals that asset correlation in the subprime market is surprisingly low (roughly 6%). By giving an intuitive and straightforward explanation for these low values, we provide valuable insight into the mechanism and evolution of the subprime crisis in general, and into the risk characteristics of a credit portfolio in particular.