In this paper we evaluate a technique to detect and recover messages from packet collisions by exploiting the capture effect. It can differentiate between collisions and packet loss and can identify the nodes involved in the collisions. This information is provided at virtually no extra cost and can produce significant improvements in existing collision mediation schemes. We characterize this technique using controlled collision experiments and evaluate it in real world flooding experiments on a 36-node sensor network.