We performed a prospective observational cohort study of the epidemiology and etiology of nosocomial pneumonia in 358 medical ICU patients in two university-affiliated hospitals. Protected bronchoscopic techniques (protected specimen brush and bronchoalveolar lavage) were used for diagnosis to minimize misclassification. Risk factors for ventilator-associated pneumonia were identified using multiple logistic regression analysis. Twenty-eight cases of pneumonia occurred in 358 patients for a cumulative incidence of 7.8% and incidence rates of 12.5 cases per 1, 000 patient days and 20.5 cases per 1,000 ventilator days. Staphylococcus aureus, Streptococcus pneumoniae, Pseudomonas aeruginosa, and Hemophilus species made up 65% of isolates from the lower respiratory tract, whereas only 12.5% of isolates were enteric gram-negative bacilli. Daily surveillance cultures of the nares, oropharynx, trachea, and stomach demonstrated that tracheal colonization preceded ventilator-associated pneumonia in 93.5%, whereas gastric colonization preceded tracheal colonization for only four of 31 (13%) eventual pathogens. By multiple logistic regression, independent risk factors for ventilator- associated pneumonia were admission serum albumin <= 2.2 g/dl (odds ratio [OR] 5.9; 95% confidence interval [CI] 2.0-17.6; p = 0.0013), maximum positive end-expiratory pressure >= 7.5 cm H2O (OR, 4.6; 95% CI, 1.4 to 15.1; p = 0.012), absence of antibiotic therapy (OR, 6.7; 95% CI, 1.8 to 25.3; p = 0.0054), colonization of the upper respiratory tract by respiratory gram-negative bacilli (OR, 3.4; 95% CI, 1.1 to 10.1; p = 0.028), pack-years of smoking (OR, 2.3 for 50 pack-years; 95% CI, 1. 2 to 4.2; p = 0.012), and duration of mechanical ventilation (OR, 3. 4 for 14 d; 95% CI, 1.5 to 7.8; p = 0.0044). Several of these risk factors for ventilator-associated pneumonia appear amenable to intervention.