Statistical methods to assess bioequivalence of a test and a reference formulation are reviewed with emphasis on the distribution of bioequivalence characteristics and the consumer risk of erroneously accepting bioequivalence. Among the procedures not exceeding a nominal consumer risk of 5%, the one with an acceptably small producer risk of erroneously rejecting bioequivalence is selected. With the exception of tmax, the following strategy is recommended: a decision in favour of bioequivalence is made if the shortest 90%-confidence interval for the ratio of the expected medians is in the bioequivalence range for the chosen characteristics of rate and extent of absorption. If the assumption of a logarithmic normal distribution is not valid, the analogous nonparametric (distribution-free) 90%-confidence interval, which is also based on the two-sample approach for the sequences reference/test and test/reference, is the procedure of choice. The issue of a modification of the bioequivalence range of 80-120% to other values for bioequivalence characteristics other than AUC (e.g. Cmax) is also addressed. Finally, a decision rule for tmax is presented.