Reports on pharmacokinetic drug–drug interaction studies account for a major fraction of manuscripts in clinical pharmacology journals; this issue of European Journal of Clinical Pharmacology contains three of them [1–3]. Most of the published reports are positive, i.e. they show some degree of change in the pharmacokinetics of drug A when drug B is co-administered. A standard conclusion of such results is that great care should be applied if the two drugs are given concomitantly. In contrast, many patients take five or more different drugs every day [4], often because there is a clear therapeutic indication for each of them, and nobody really cares about interactions as such, but rather a general clinical monitoring of adverse events and therapeutic effects is done. Indeed, this turns out to be sufficient for patient safety in most cases. However, if in clinical studies a pronounced pharmacokinetic interaction between two partners has been reported, such as the effect of gemfibrozil on cerivastatin [5] or that of mibefradil on CYP3A4 substrates [6], this is also often disregarded by clinicians, and these two cases even required market withdrawal of the drugs. How may we estimate true risk of a drug–drug interaction and identify those interactions that are clinically important in order to generate a signal which clinicians can rely on? Of course, the most reliable data would come from prospective, randomised, double-blind clinical trials using clinical endpoints, but this approach is not feasible for the multitude of potential interactions. We therefore rely on pharmacokinetics and/or on pharmacodynamics of drug A. This is justified because most interactions on the pharmacokinetic level will just change the intensity of the effect and will not qualitatively change therapy, although there may be exceptions in the case of an altered formation of active metabolites and/or a different tissue distribution. How much of a pharmacokinetic change should then be considered as relevant? The definition of bioequivalence, which is considered as the basis for interchangeability of two preparations, has been strictly formalized in respective guidelines [7–10] and usually requires that two preparations differ in mean area under curve (AUC) and Cmax values by not more than 0.8to 1.25-fold. Interchangeability is also the question in the case of pharmacokinetic drug–drug interactions: The clinician needs to know whether the concentrations and related effects of drug A are the same with and without concomitant administration of drug B, or whether some adjustment of the treatment or dosage is needed. Accordingly, it has been proposed to handle drug– drug interactions as an equivalence problem [11]. From a statistical point of view, this is very reasonable because it addresses power limitations and the validity of negative results, and there are also tools to assess interindividual variation in the degree of change (“individual bioavailability” methods) [8, 10]. In contrast to bioequivalence, however, the margins defining that a treatment with co-medication is not interchangeable to the treatment without co-medication are less clear. To be on the safe side, the same strict limits as applied for bioequivalence assessment have been proposed, and fulfilling the standard bioequivalence criteria usually means that a drug–drug interaction is not clinically relevant. However, even in regulatory guidelines, it is acknowledged that depending on the drug, different “no effect boundaries” may be appropriate [12, 13]. The criteria for the assessment of bioequivalence and of drug–drug interactions differ because Eur J Clin Pharmacol (2007) 63:897–899 DOI 10.1007/s00228-007-0357-6