OBJECTIVES: The working alliance predicts improvement following general psychotherapy, but how it operates in brief interventions conducted with medically ill patients is unknown. Also, the role of the working alliance may differ in emotion-focused versus educational interventions. METHODS: We report secondary analyses of a randomized clinical trial (Keefe et al.) [35], in which patients with rheumatoid arthritis (RA) received four nurse-provided sessions of either a) Clinician-assisted Emotional Disclosure (CAED), which emphasized the disclosure, expression, and processing of emotions related to stressful events; or b) Arthritis Education (AE), which provided basic education about RA. The Working Alliance Inventory was completed by both patient and nurse after each session. Patients were evaluated on multiple health measures at baseline and 1, 3, and 12months post-treatment. RESULTS: Analyses compared the alliance between interventions and related the alliance to outcomes within interventions. Patients in CAED reported a lower alliance than patients in AE. Interestingly, in CAED, lower alliance ratings predicted better outcomes (improved functioning, lower pain behaviors, lower inflammation, lower daily stress), whereas in AE, the working alliance was largely not predictive of outcomes. CONCLUSION: Having nurses encourage emotional disclosure among patients with RA reduced the patients' working alliance, but a lower alliance nonetheless predicted better patient outcomes, perhaps reflecting successful engagement in an intervention that is emotionally and relationally challenging. The level and predictive validity of the working alliance likely depends on patient, provider, and intervention factors, and further study of the working alliance in psychosocial interventions in the medical context is needed.