Objective. To determine rates and risk factors for self-discharge by Aboriginal medical inpatients at Alice Springs Hospital. Methods. Prospective cohort study. Interviews were conducted in primary language by Aboriginal Liaison Officers, from July 2006 to August 2007. Topics included understanding of diagnosis, satisfaction with services and perceptions of staff and environment. Risk factors for self-discharge were then determined prospectively. Results. During the study period 202 (14.7%) of 1380 patients admitted to general medical units at Alice Springs Hospital, were interviewed. Self-discharge rates for all admissions were significantly lower during the study period than they had been previously (pre-study, mean 22.9±standard error 0.3%; study, 17.0±0.2%) (P<0.001). Most interviewees (73.4%) did not know their reason for admission (73.4%) or estimated length of stay (82.3%). Forty interviewees (19.8%) self-discharged. Mean monthly self-discharge rates differed between the three medical units (Unit A, 13.9±0.3%; Unit B, 17.3±1.37%; Unit C, 20.0±0.4%) (P≤0.005). Multivariable predictors of self-discharge included male sex (hazard ratio (HR) 2.4; 95% confidence interval (CI) 1.1, 5.2), a past history of self-discharge (HR 3.2; 95%CI 1.5, 6), planned transfer to a tertiary referral centre (HR 3.8; 95%CI 1.3-7.4) and a desire to drink alcohol (HR 4.5; 95%CI 1.8-10.2). Conclusions. Physician, institutional and patient factors all contribute to self-discharge. Improving cultural safety may be the key to lowering self-discharge rates. What is known about the topic? Rates of self-discharge by Aboriginal adults in Central Australia are the highest reported worldwide. Previous studies have been retrospective and focussed on patient demographics without addressing the environmental and cultural contexts in which self-discharge occurs. What does this paper add? In this acute care setting, we found a pervasive failure to communicate effectively with Aboriginal patients. Consequently, most patients were unaware of their diagnosis or length of stay. Self-discharge was a common practice; nearly half of all previously admitted patients had self-discharged in the past. We demonstrate that physician, hospital and patient factors all contribute to this practice. Prospectively determined risk factors included the treating medical team, the need for transfer outside Central Australia, and patient factors such as male gender and alcohol dependence. Self-discharge rates fell significantly with Aboriginal Liaison involvement. What are the implications for practitioners? Cross-cultural communication skills must be markedly improved among medical staff caring for this marginalised population. Critical to reducing rates of self-discharge are improvements in institutional cultural safety by involving Aboriginal Liaison Officers and family members. However, persistently high self-discharge rates suggest a need to redirect medical services to a more culturally appropriate community-based model of care.