Among critically ill patients, acute kidney injury (AKI) is a relatively common complication that is associated with an increased risk for death and other complications. To date, no treatment has been developed to prevent or attenuate established AKI. Dialysis often is required, but the optimal timing of initiation of dialysis is unknown. Data from the Program to Improve Care in Acute Renal Disease (PICARD), a multicenter observational study of AKI, were analyzed. Among 243 patients who did not have chronic kidney disease and who required dialysis for severe AKI, we examined the risk for death within 60 d from the diagnosis of AKI by the blood urea nitrogen (BUN) concentration at the start of dialysis (BUN < or = 76 mg/dl in the low degree of azotemia group [n = 122] versus BUN > 76 mg/dl in the high degree of azotemia group [n = 121]). Standard Kaplan-Meier product limit estimates, proportional hazards (Cox) regression methods, and a propensity score approach were used to account for selection effects. Crude survival rates were slightly lower for patients who started dialysis at higher BUN concentrations, despite a lesser burden of organ system failure. Adjusted for age, hepatic failure, sepsis, thrombocytopenia, and serum creatinine and stratified by site and initial dialysis modality, the relative risk for death that was associated with initiation of dialysis at a higher BUN was 1.85 (95% confidence interval 1.16 to 2.96). Further adjustment for the propensity score did not materially alter the association (relative risk 1.97; 95% confidence interval 1.21 to 3.20). Among critically ill patients with AKI, initiation of dialysis at higher BUN concentrations was associated with an increased risk for death. Although the results could reflect residual confounding by severity of illness, they provide a rationale for prospective testing of alternative dialysis initiation strategies in critically ill patients with severe AKI.