The application of pharmacogenetics holds great promise for individualized therapy. However, it has little clinical reality at present, despite many claims. The main problem is that the evidence base supporting genetic testing before therapy is weak. The pharmacology of the drugs subject to inherited variability in metabolism is often complex. Few have simple or single pathways of elimination. Some have active metabolites or enantiomers with different activities and pathways of elimination. Drug dosing is likely to be influenced only if the aggregate molar activity of all active moieties at the site of action is predictably affected by genotype or phenotype. Variation in drug concentration must be significant enough to provide "signal" over and above normal variation, and there must be a genuine concentration-effect relationship. The therapeutic index of the drug will also influence test utility. After considering all of these factors, the benefits of prospective testing need to be weighed against the costs and against other endpoints of effect. It is not surprising that few drugs satisfy these requirements. Drugs (and enzymes) for which there is a reasonable evidence base supporting genotyping or phenotyping include suxamethonium/mivacurium (butyrylcholinesterase), and azathioprine/6-mercaptopurine (thiopurine methyltransferase). Drugs for which there is a potential case for prospective testing include warfarin (CYP2C9), perhexiline (CYP2D6), and perhaps the proton pump inhibitors (CYP2C19). No other drugs have an evidence base that is sufficient to justify prospective testing at present, although some warrant further evaluation. In this review we summarize the current evidence base for pharmacogenetics in relation to drug-metabolizing enzymes.