A large-scale online teacher resource exchange is studied to examine the ways in which metadata influence teachers' selection of resources. A hierarchical linear modeling approach was used to tease apart the simultaneous effects of resource features and author features. From a decision heuristics theoretical perspective, teachers appear to rely on complex heuristics that integrate many dimensions when determining whether to download a resource. Most surprisingly, numbers of ratings more strongly predict downloads than do mean rating levels, such that multiple negative ratings appear to attract more downloads than do few positive ratings. Implications for system design are discussed.