Some researchers assert that “normal” mixing of words from one language into another occurs more often in multilingual individuals with aphasia (PWA) than in matched healthy controls (HC) (e.g., Chengappa et al., 2004). In one case-study, language-mixing was seen to help PWA increase their communication (Sebastian et al., 2012). Our discourse experiment compared lexical language-mixing in bilingual PWAs and matched HCs to determine its effects on discourse length and quality. Participants Our participants were 8 Kannada-English speakers, 5 mild to moderate non-fluent aphasia (according to the WAB) (X = 36.4 months post-onset;) and 3 matched controls (Mean age of participants: 57.83; mean years of education: 14.76). Methods Each participant recounted three personal narratives and described three pictures in three language conditions. In the first condition participants were shown a photo of a Kannada-speaker and told in recorded Kannada that the assigned-listener spoke only Kannada, so they should speak only Kannada. The second condition was for English, mutatis mutandis. In the third condition participants heard, in appropriately language-mixed instructions, that the assigned listener was bilingual so they should use both languages. Transcription and Scoring All discourse recordings were transcribed by two bilingual Kannada-English SLPs, inter-rater reliability was assured. Total words per narrative, percent of language-mixed words, and ratings of ‘overall success’ per narrative (1: extremely poor; 7: excellent) were analyzed. Results Overall, the PWA produced fewer total words (X = 43.21) than the HC (X = 87.58). The HC produced most words in the Kannada-only condition (98.83 vs. 83.16 in the English-only condition and 80.8 in the ok-to-mix condition). The PWA, by contrast, produced most words in the ok-to-mix condition, especially in the personal narratives where they produced many more words in the mixed-language condition (68.8) than in the English-only condition (31.6). Mixing occurred most often in the personal narratives of the PWA: 14% of the time vs. 4.32% for the PWA for picture description; for HC, 2.54% and 2.12% respectively. However the greater amount of mixing PWA included in the personal narratives was not associated with marked differences in overall narrative success for the personal narratives (3.77) and the picture descriptions (3.37). Conclusion In sum, while PWA may not altogether avoid language-mixing with monolingual listeners, they appear to mix less when they know their listeners are monolinguals, not bilinguals. PWAs’ personal narratives are markedly longer when their listener is bilingual, though ratings of ‘overall success’ of the discourse are not linked to the degree of patients’ mixing. The lack of differences in the rated success of the more- and less-mixed samples suggests that, contrary to our hypothesis, language-mixing did not improve quality of communication.