Clinicians making treatment decisions are faced with ever-growing numbers of therapies, each supported by different types of clinical data. By bringing together large amounts of data, meta-analysis has emerged as a useful tool for generating hypotheses with which to plan definitive trials, and it has also been recommended as a basis for decision making in the absence of definitive trials. In several instances, early meta-analyses have provided evidence of efficacy that was subsequently confirmed. However, in other instances, the results of initial meta-analyses have disagreed with the results of subsequent large-scale trials. Nitrate and magnesium therapy for acute myocardial infarction are two contemporary examples of treatments about which hypothesis-generating meta-analyses and subsequent large trials have disagreed. We review the issues surrounding the interpretation of meta-analyses in these cases, and we suggest that the appropriate use of meta-analyses in clinical decision making be carefully placed in the context of a review of pathophysiologic principles and the results of basic laboratory research and individual trials.