Ventilator-associated pneumonia (VAP) is a common and serious nosocomial infection. Accurate, timely diagnosis enables affected patients to receive appropriate therapy and avoids mistreatment of patients having other conditions. To review the published medical literature describing the precision and accuracy of clinical, radiographic, and laboratory data to diagnose bacterial VAP relative to a histological gold standard. English-language articles identified by a structured search strategy using MEDLINE (January 1966-October 31, 2006) and Google Scholar. Additional articles were identified through the reference lists of studies and review papers identified by the search strategy. Included studies described clinical findings associated with VAP in 25 or more patients receiving mechanical ventilation who subsequently underwent pulmonary biopsy or autopsy. Fourteen studies describing clinical findings in 655 patients met inclusion criteria. Data were abstracted onto a structured form, allowing calculation of the likelihood ratios (LRs) for each sign or combination of findings. The presence or absence of fever, abnormal white blood cell count, or purulent pulmonary secretions do not substantively alter the probability of VAP. However, the combination of a new radiographic infiltrate with at least 2 of fever, leukocytosis, or purulent sputum increases the likelihood of VAP (summary LR, 2.8; 95% confidence interval, 0.97-7.9). The absence of a new infiltrate on a plain chest radiograph lowers the likelihood of VAP (summary LR, 0.35; 95% confidence interval, 0.14-0.87). Fewer than 50% neutrophils on cell count analysis of lower pulmonary secretions makes VAP unlikely (LR range, 0.05-0.10). Routine bedside evaluation coupled with radiographic information provides suggestive but not definitive evidence that VAP is present or absent. Given the severity of VAP and the frequency of serious conditions that can mimic VAP, clinicians should be ready to consider additional tests that provide further evidence for VAP or that establish another diagnosis.