Accelerated stability data obtained at a single temperature (40 degrees C) is statistically evaluated and the utility of the actual value of remaining drug content observed in the test for estimation of shelf-life is discussed. The distribution of the time required for 10% degradation at 25 degrees C, t90(25), is calculated from the observed drug content, taking the experimental errors into consideration. The probability that the t90(25) is longer than a specified shelf-life, which is calculated from the distribution, is represented as a function of the observed drug content. The probability was found to depend on the experimental errors in the assay and temperature control as well as in the activation energy of degradation (Ea). The relationship between the observed drug content and the shelf-life assured with a certain probability is shown graphically, so that one can reasonably predict the expected shelf-life as a function of Ea if the accuracy and precision of the assay and temperature control are known.