Hypoalbuminemia predicts death in dialysis patients. Although hypoalbuminemia has been attributed to malnutrition, evidence of inflammation (C-reactive protein [CRP] and cytokine levels) has recently been recognized to predict albumin concentration in dialysis patients. We measured CRP and albumin levels in October 1995 in 91 hemodialysis (HD) patients. During a 34-month follow-up period, we determined the incidence and cause of death. Patients were divided into four groups based on serum albumin levels (<3.5 [lowest quartile], 3.5 to 3.8, 3.9 to 4.0, and >4.0 g/dL [highest quartile]). Survival differed among the four groups (P = 0.0063). Patients with albumin levels greater than 4.0 g/dL had the greatest survival. Kaplan-Meier survival estimates of patients from varying CRP quartiles (<2.6, 2.6 to 5.2, 5.3 to 11.5, and >11.5 microg/mL) differed among the four groups (P < 0.0001). The group with the greatest CRP level (>11.5 microg/mL) had the lowest survival. Multivariate analysis using the Cox proportional hazards model showed that only CRP level (chi-square = 21.11; P < 0.0001) and age (chi-square = 5.44; P = 0.020) predicted death. Albumin level (chi-square = 0.16; P = 0.69) was not predictive. Only when CRP was excluded from the model did low serum albumin level (chi-square = 12. 04; P = 0.0004) predict death. CRP level (chi-square = 16.79; P < 0. 0001) and age (chi-square = 6.38; P = 0.012) also superceded albumin level (chi-square = 0.45; P = 0.51) in predicting cardiovascular mortality. Although values for blood urea nitrogen, creatinine, and normalized protein catabolic rate were significantly less among patients who died, these parameters, as well as cholesterol level and diabetes, were not important predictors of death in multivariate analysis. The acute-phase response or the cause of the acute-phase response is largely responsible for the effect of hypoalbuminemia on mortality in HD patients.