In 1992, the US Food and Drug Administration declared that two drugs demonstrate average bioequivalence (ABE) if the log-transformed mean difference of pharmacokinetic responses lies in (−0.223, 0.223). The most widely used approach for assessing ABE is the two one-sided tests procedure. More specifically, ABE is concluded when a 100(1 − 2α) % confidence interval for mean difference falls within (−0.223, 0.223). As known, bioequivalent studies are usually conducted by crossover design. However, in the case that the half-life of a drug is long, a parallel design for the bioequivalent study may be preferred. In this study, a two-sided interval estimation — such as Satterthwaite's, Cochran–Cox's, or Howe's approximations — is used for assessing parallel ABE. We show that the asymptotic joint distribution of the lower and upper confidence limits is bivariate normal, and thus the sample size can be calculated based on the asymptotic power so that the confidence interval falls within (−0.223, 0.223). Simulation studies also show that the proposed method achieves sufficient empirical power. A real example is provided to illustrate the proposed method.