BACKGROUND: Clinician stress is common, but few studies have examined its relationship with communication behaviors. OBJECTIVE: To investigate associations between clinician stress and patient-clinician communication in primary HIV care. DESIGN: Observational study. PARTICIPANTS: Thirty-three primary HIV clinicians and 350 HIV-infected adult, English-speaking patients at three U.S. HIV specialty clinic sites. MAIN MEASURES: Clinicians completed the Perceived Stress Scale, and we categorized scores in tertiles. Audio-recordings of patient-clinician encounters were coded using the Roter Interaction Analysis System. Patients rated the quality of their clinician's communication and overall quality of medical care. We used regression with generalized estimating equations to examine associations between clinician stress and communication outcomes, controlling for clinician gender, clinic site, and visit length. KEY RESULTS: Among the 33 clinicians, 70 % were physicians, 64 % were women, 67 % were non-Hispanic white, and the mean stress score was 3.9 (SD 2.4, range 0-8). Among the 350 patients, 34 % were women, 55 % were African American, 23 % were non-Hispanic white, 16 % were Hispanic, and 30 % had been with their clinicians >5 years. Verbal dominance was higher for moderate-stress clinicians (ratio = 1.93, p < 0.01) and high-stress clinicians (ratio = 1.76, p = 0.01), compared with low-stress clinicians (ratio 1.45). More medical information was offered by moderate-stress clinicians (145.5 statements, p <0.01) and high-stress clinicians (125.9 statements, p = 0.02), compared with low-stress clinicians (97.8 statements). High-stress clinicians offered less psychosocial information (17.1 vs. 19.3, p = 0.02), and patients of high-stress clinicians rated their quality of care as excellent less frequently than patients of low-stress clinicians (49.5 % vs. 66.9 %, p < 0.01). However, moderate-stress clinicians offered more partnering statements (27.7 vs. 18.2, p = 0.04) and positive affect (3.88 vs. 3.78 score, p = 0.02) than low-stress clinicians, and their patients' ratings did not differ. CONCLUSIONS: Although higher stress was associated with verbal dominance and lower patient ratings, moderate stress was associated with some positive communication behaviors. Prospective mixed methods studies should examine the complex relationships across the continuum of clinician well-being and health communication.