Nine experiments involving young adults (N = 525) tested the roles of local (sentence) and global (discourse) contexts on lexical processing. Contextual material was presented auditorily, and naming times for the last (visually presented) word were collected. Experiment 1 tested the local contexts alone and found facilitation of naming latencies when local contexts were related to the target word. Subsequent experiments, using Varying baseline conditions, found that globally related material affected naming latency in all cases, whereas the same locally related material that was used in the first study now had no facilitation effect. The globally related material had an immediate effect on naming times. The authors argue that the results are inconsistent with associatively based models and with various hybrid models of context effects and that a discourse-based model best accounts for the data.