OBJECTIVE: Starling's equation indicates that reduced oncotic pressure gradients will favor edema formation, and the current consensus definition of acute respiratory distress syndrome (ARDS) excludes only the hydrostatic pressure contribution. We hypothesized that low serum total protein levels might correlate with the likelihood of ARDS in at-risk patients because serum total protein is the chief determinant of oncotic pressure in humans. DESIGN: Regression analysis to compare outcomes in patients with low serum total protein levels with outcomes in patients with normal serum total protein levels with respect to weight change, development of ARDS, and mortality. SETTING: Intensive care units (ICUs) of seven clinical centers in North America. PATIENTS: A total of 455 ICU patients who met consensus criteria for severe sepsis (178 of whom developed ARDS) from a recently completed prospective clinical trial. INTERVENTION: None. MEASUREMENTS AND MAIN RESULTS: We found that 92% of the patients developing ARDS had low or borderline serum total protein levels (<6 g/dL). Logistic and multiple regression analyses confirmed that of 18 clinical variables, initial serum total protein level and protein change over time were the most statistically significant predictors of weight gain, prolonged mechanical ventilation, ARDS development, and mortality in the study population. This correlation remained significant after adjustment for the other major predictors of outcome present at baseline (ie, Acute Physiology and Chronic Health Evaluation II score). CONCLUSIONS: Hypoproteinemia is significantly correlated with fluid retention and weight gain, development of ARDS and poor respiratory outcome, and mortality in patients with sepsis. Prospective, randomized trials of serum protein manipulation are needed to establish whether there is a cause-effect relationship to this association.