PURPOSE: Semantic feature analysis (SFA) was used to determine whether training contextually related words would improve the discourse of individuals with nonfluent aphasia in preselected contexts. METHOD: A modified multiple-probes-across-behaviors design was used to train target words using SFA in 3 adults with nonfluent aphasia. Pretreatment, posttreatment, and follow-up sessions obtained language samples for the preselected contexts. Contexts included 4 story retellings and 4 procedure explanations. RESULTS: All participants improved naming ability for treated words. No generalization to untrained items was found. Within discourse samples, participants increased number of target words produced from pretreatment to posttreatment sessions but did not increase lexical diversity across samples. Participants maintained performance on standardized measures from the beginning to the end of the study. CONCLUSIONS: Results support and extend previous research by indicating that SFA improves confrontational naming ability and may benefit word retrieval in discourse production of closed-set contexts.