Background Substantial variability in hemoglobin levels has been associated with increased mortality risk in hemodialysis patients. Variability also has been associated with concurrent comorbid conditions and hospitalization. Adequate adjustment for confounding by disease severity is needed to estimate the association of hemoglobin level variability with mortality risk. Study Design Retrospective cohort study. Setting & Participants Medicare hemodialysis patients in 3 groups: prevalent on July 1, 2006 (n = 133,246), prevalent on July 1, 1996 (n = 78,602), and incident between January 1, 2005, and June 30, 2006 (n = 24,999). Predictor Hemoglobin level variability estimated using the residual deviation around the linear trend in hemoglobin levels during a 6-month entry period. Outcome Time to death. Measurements We fit Cox models of 1-year mortality with and without adjustment for disease severity (comorbid conditions, hospitalization days, and months with hemoglobin level <10 g/dL), measured concurrently with hemoglobin level variability. Results Disease severity was associated positively with hemoglobin level variability in all groups. Before adjustment for disease severity, HRs for hemoglobin level variability were 1.27 (95% CI, 1.24-1.31) per 1 g/dL for patients prevalent in 2006, 1.32 (95% CI, 1.27-1.38) for patients prevalent in 1996, and 1.08 (95% CI, 1.03-1.13) for patients incident in 2005-2006. After adjustment, HRs for hemoglobin level variability were 1.02 (95% CI, 0.99-1.05), 1.07 (95% CI, 1.03-1.12), and 1.01 (95% CI, 0.95-1.06), respectively. Limitations We did not adjust for time-varying confounding of hemoglobin level; an inclusion requirement introduces potential selection bias; our findings may not apply to incident hemodialysis patients younger than 65 years; assessment of comorbid conditions from claims is subject to misclassification, with possible residual confounding attributable to comorbid conditions; this observational study cannot prove causality. Conclusions After adjustment for concurrent disease severity, evidence supporting an association between hemoglobin level variability and mortality risk was weak and inconsistent. The clinical utility of hemoglobin level variability may be limited.