All pharmaceutical products are required by law to display the shelf-life on the packaging, which indicates the period during which the consumer can expect the product to be safe and effective. Standard methods for determining the label shelf- life from stability data are limited to linear models. Where a number of batches are used to determine a label shelf-life, the current regulatory method (unintentionally) penalizes good statistical design. In this work it is shown that the current approaches may not provide the consumer with the desired level of confidence that the product is safe and effective. Consequently, a new definition for the label shelf-life is proposed, such that the consumer can be confident that a certain percentage of the product will meet the specification by the expiry date. Several methods for obtaining such a label shelf-life under linear model and generalized linear model assumptions are proposed and evaluated using simulation studies. The definition is extended so that a single shelf-life can be obtained from multiple batches.