Demand for dialysis for patients with end-stage renal disease is growing, as is the comorbidity of dialysis patients. Accurate prediction of those destined to die quickly despite dialysis could be useful to patients, providers, and society in making decisions about starting dialysis. To determine whether age and comorbidity accurately predict death within 6 months of first dialysis for end-stage renal disease, a prospective cohort study of 822 patients starting dialysis at one of 11 Canadian centers was performed. Patient characteristics were recorded at first dialysis. Follow-up continued until death or study end (at least 6 months after enrollment). One hundred thirteen of 822 (13.7%) patients died within 6 months. Although an existing scoring system predicted prognosis, adverse scores greater than 9 were found in only 9.7% of those who died; only 52% of those who scored higher than 9 died within 6 months. No score cutoff point combined high true-positive and low false-positive rates for predicting early death. Age, severity of heart failure or peripheral vascular disease, arrhythmias, malnutrition, malignancy, or myeloma were independent prognostic factors identified in multivariate models. However, the best fit discriminant and logistic models were also unable to accurately predict death within 6 months. Clinicians were very accurate in assigning patients to prognostic groups up to a 50% risk of death by 6 months, above which they tended to overestimate risk. However, clinicians were only marginally better than the predictive models in determining whether a given high-risk patient would die. The inability of a scoring system or clinical intuition to accurately predict death soon after starting dialysis for end-stage renal disease suggests that limiting access to dialysis on the basis of likely short survival may be inappropriate in Canada.