In an attempt to determine whether a generic product is bioequivalent to its reference listed drug (RLD), the comparison of the test and reference distributions of a pharmacokinetic parameter is necessary. Usually, the response variable is a logarithmically transformed bioequivalence metric like the area under the curve (AUC), or the maximum concentration (Cmax) and the statistical test consists of comparing the average responses from the test and reference distributions. For locally-acting dosage forms like creams, gels and ointments, the available data often follow a paired-sample design where a subject is exposed to both the Test and the Reference formulations and the appropriate criterion for determining bioequivalence, is the two one-sided (TOST) confidence interval. Such approaches, although theoretically correct, cannot be always applied as the special nature of the data introduces challenges like the limited number of available subjects and/ or their replicates, the unusually high within-subject and between-subject variability and the presence of outlying subjects. Concerns arising from the use of a bioequivalence criterion are related to the sensitivity of the test for detecting only meaningful differences and at the same time, not rejecting good products. Additionally, statistical power is affected by various factors like the bioequivalence limit, the sample size, the within-reference variability and the choice of the regulatory constants. These concerns are going to be discussed in reference to data coming from an in-vitro permeation test, using human skin. The data will be analyzed using regular average bioequivalence and different forms of reference-scaled average bioequivalence and the advantages and imposed risks of each approach will be discussed.