To evaluate the accuracy of clinical judgment in the diagnosis and treatment of nosocomial pneumonia in ventilated patients, we studied 84 patients suspected of having nosocomial pneumonia because of the presence of a new pulmonary infiltrate and purulent tracheal secretions. We prospectively evaluated the accuracy of diagnostic predictions and therapeutic plans independently formulated by a team of physicians aware of all clinical, radiologic and laboratory data, including the results of Gram-stained bronchial aspirates. Definite (n = 51) or probable (n = 33) diagnoses could be established in all patients by strict histopathologic and/or bacteriologic criteria. Only 27/84 patients were diagnosed as having pneumonia. Organisms responsible for pneumonias were identified by quantitative cultures of samples obtained using a protected specimen brush or pleural fluid cultures. Four hundred eight predictions were made for the 84 studied patients. Clinical diagnoses for patients subsequently diagnosed as having pneumonia were accurate in 81/131 cases (62 percent). Furthermore, only 43/131 (33 percent) therapeutic plans proposed for these patients represented effective therapy. Common causes of inappropriate treatment included failure to diagnose pneumonia (50 plans), failure to effectively treat highly resistant organisms (21 plans), and failure to treat all organisms in cases of polymicrobial pneumonia (14 plans). Therapeutic plans formulated for patients without pneumonia included the unnecessary use of antibiotics in 45/277 cases (16 percent). These findings indicate that the use of clinical criteria alone does not permit the accurate diagnosis of nosocomial pneumonia in ventilated patients, and commonly results in inappropriate or inadequate antibiotic therapy for these patients.