The main objective of the stability study of pharmaceutical products is to estimate the shelf life indicated on the container label. For the determination of the label shelf life the International Conference on Harmonisation (ICH) Guideline recommends the confidence interval approach, but its adequacy is questionable. Instead of the current regulation, several proposals are known in the literature applying the tolerance interval approach. The aim of our paper is to extend the tolerance interval model assuming two distinct influence sources, namely the tablet-to-tablet variability and the analytical measurement error. Another main concept in the paper is that the tolerance interval is constructed for the true (and not for the measured) assay content. Investigating the influence of several factors on the width of the tolerance interval, a simulation is performed applying different number of samples and repeated analysis, as well as different variance components regarding the tablet-to-tablet variability and the analytical measurement error. An example is also given to illustrate the difference between the former and the proposed tolerance interval models and to compare them with the present regulation of the ICH. Copyright © 2010 John Wiley & Sons, Ltd.