Recent work has explored the idea of using trust networks to supplement ratings information in community-based information systems, including algorithms to infer missing values in the trust network. Current trust inference algorithms sometimes make undesirable inferences because they do not fully use information about distrust and sometimes make inferences based on weak support. Further, many algorithms do not consider the problem of trust scope, where one may trust someone's opinions about movies but not books. We present GePuTTIS, a trust inference system that reasons about support levels, distrust, and trust scope. We demonstrate that it improves prediction performance in a collaborative filtering dataset.