Ventilator-associated pneumonia (VAP) rates are advocated as a measure of hospitals' quality of care for critically ill patients. The standard definition used to measure VAP rates, however, is constructed of nonspecific clinical signs common to many common complications of critical care. We created a model in which we estimated the probability of patients with 6 different complications of critical care fulfilling diagnostic criteria for VAP. We then calculated how the apparent prevalence of VAP varies depending on the prevalence of these other conditions in an intensive care unit. Despite keeping the true, underlying prevalence of VAP fixed at 10%, the apparent rate of VAP varied between 6.0% and 31.6%, depending on the prevalence of other conditions. The addition of microbiological criteria to standard clinical criteria decreased the range of apparent VAP to 3.5%-15.5%. These wide margins of variability suggest that VAP rates are an unreliable measure of quality of care.