Purpose: To evaluate the characteristics and outcomes of critically ill patients with severe acute kidney injury (AKI) treated and not treated with renal replacement therapy (RRT). Methods: Secondary analysis of a multi-centre cohort study. Primary exposure was RRT. Primary outcome was propensity and multi-variable adjusted-hospital mortality. Results: We studied 1250 patients (71.3%) who received and 502 (28.7%) who did not receive RRT. Reasons for not starting RRT (not mutually exclusive) were limitations of support (33.6%, n = 169), adequate urine output (46.2%; n = 232), plan to observe (56.4%; n = 283), and advanced age (12.6%; n = 63). Mortality was higher in those not receiving RRT due to limitations and advanced age but lower for adequate urine output and plan to observe. Propensity and multi-variable adjusted analysis showed no statistical difference in hospital mortality (adj-OR 1.47; 95% CI, 0.93-2.24) in patients receiving RRT. Results were similar in a sensitivity analysis restricted to patients fulfilling risk, injury, failure, loss, end-stage kidney disease-FAILURE criteria (37.0%; n = 446) (adj-OR 1.36; 95% CI, 0.70-2.66). Conclusion: In this cohort, reasons for not starting RRT included limitations of support and perception of impending renal recovery. Despite similar risk of mortality after adjusting for selection bias and confounders, RRT-treated patients were fundamentally different from non-treated patients across a spectrum of variables that precludes valid comparison in observational data. © 2013 Elsevier Inc.