This article examines the usefulness of meta-analysis, and articulates many of the criticisms that have been made of its workings. An attempt is made to outline the precautions that have to be taken before a scientifically useful and meaningful meta-analysis can be carried out. The problems encountered include heterogeneity of samples, conditions, interventions and end-points; narrow focus; curvilinearity of regression; lack of independence of determinants; synergistic interactions; contradictory experimental results. It is suggested that best-evidence synthesis, or theory-directed analysis, might be a safer option.