The effect of different lot-to-lot variability levels on the prediction of stability are studied based on two statistical models for estimating degradation in real time and accelerated stability tests. Lot-to-lot variability is considered as random in both models, and is attributed to two sources-variability at time zero, and variability of degradation rate. Real-time stability tests are modeled as a function of time while accelerated stability tests as a function of time and temperatures. Several data sets were simulated, and a maximum likelihood approach was used for estimation. The 95% confidence intervals for the degradation rate depend on the amount of lot-to-lot variability. When lot-to-lot degradation rate variability is relatively large (CV > or = 8%) the estimated confidence intervals do not represent the trend for individual lots. In such cases it is recommended to analyze each lot individually.