'Highly variable drugs' have been defined as those drugs for which the within-subject variability (WSV) equals or exceeds 30% of the maximum concentration (Cmax) and/or the area under the concentration versus time curve (AUC). Despite the fact that highly variable drugs are generally safe with flat dose response curves, the bioequivalence of their formulations is a problem because the high variability means that large numbers of subjects are required to give adequate statistical power. Highly variable drug products are poor quality formulations where high within-formulation variability (e.g. tablet to tablet variability) poses a problem rather than high innate WSV of the drug itself. A further problem caused by high variability is that a subset of the population may respond differently to the two formulations producing a significant subject x formulation interaction. Practical examples are shown using replicate designs. The methods proposed to deal with the problems posed by highly variable drugs include: (i) Drug regulatory jurisdictions states that the 90% confidence interval (90% CI) around the test to reference geometric mean ratio (GMR) is required to fit with bioequivalence acceptance limits of 0.8 - 1.25 for both Cmax and AUC. The WSV for single point estimation of Cmax is often greater than that for AUC. One strategy therefore is not to require a 90% CI for Cmax of drugs that do not exhibit a toxicity associated with Cmax and merely require the GMR to fall within the acceptance limits. (ii) To arbitrarily broaden the bioequivalence acceptance limits. For example, to permit a sponsor to justify the use of wider limits e.g the 90% CI around the GMR of Cmax values might be required to fit within acceptance limits of 0.75 - 1.33 or even 0.70 - 1.42. (iii) A more systematic approach would be to broaden the acceptance limits by scaling to either the residual variance from a 2-period design or to the WSV of the reference product in a replicate design. Subsequent evaluations of scaling procedures have demonstrated that smaller numbers of subjects are required for bioequivalence studies on formulations of highly variable drugs. A disadvantage of scaling is that the method is less sensitive to differences between the means compared with unscaled treatment, such that the GMR may prove to be unacceptably low or high. This possibility has let to a suggestion that the GMR must fall within acceptance limits of 0.8 - 1.25 in scaled treatments. (iv) A similar method is to scale the metric rather than the acceptance limits. This method was proposed by the United States' Food and Drug Administration in the context of Individual bioequivalence, but may also be applied (v) to average bioequivalence. (vi) To carry out bioequivalence studies at steady state whenever a multiple dose regimen is ethically acceptable for healthy volunteers. This solution is based on the observation that high variability in a single dose study tends to be dampened at steady state, thus increasing statistical power. Drug regulators have not favored this approach on the grounds that bioequivalence testing should be based on the most discriminating test possible. (vii) Finally the use of metabolite data has been proposed since in many (but by no means all) cases, metabolite is less highly variable than that of the parent drug. This subject remains controversial except when the administered substance is a prodrug which converted by metabolism into the active drug.