In the intention-to-treat analysis of the Hemodialysis Study, all-cause mortality did not differ significantly between the high versus standard hemodialysis dose groups. The association of mortality with delivered dose within each of the two randomized treatment groups was examined, and implications for observational studies were considered. Time-dependent Cox regression was used to relate the relative risk (RR) for mortality to the running mean of the achieved equilibrated Kt/V (eKt/V) over the preceding 4 mo. eKt/V was categorized by quintiles within each dose group. Analyses were controlled for case-mix factors and baseline anthropometric volume. Within each randomized dose group, mortality was elevated markedly when achieved eKt/V was in the lowest quintile (RR, 1.93; 95% confidence interval [CI], 1.40 to 2.66; P < 0.0001 in the standard-dose group; RR, 2.04; 95% CI, 1.50 to 2.76; P < 0.0001 in the high-dose group; RR relative to the middle quintiles). The mortality rate in the lowest eKt/V quintile of the high-dose group was higher than in the full standard-dose group (RR, 1.59; 95% CI, 1.29 to 1.96; P < 0.0001). Each 0.1 eKt/V unit below the group median was associated with a 58% higher mortality in the standard-dose group (P < 0.001) and a 37% higher mortality in the high-dose group (P < 0.001). The magnitude of these dose-mortality effects was seven- to 12-fold higher than the upper limit of the 95% CI from the intention-to-treat analysis. The effects were attenuated in lagged analyses but did not disappear. When dialysis dose is targeted closely, as under the controlled conditions of the Hemodialysis Study, patients with the lowest achieved dose relative to their target dose experience markedly increased mortality, to a degree that is not compatible with a biologic effect of dose. The possibility of similar (albeit smaller) biases should be considered when analyzing observational data sets relating mortality to achieved dose of dialysis.