Diagnostic technologies for rolling element bearings are relatively well developed, but accurate prediction of remaining life once an incipient fault has been detected is considerably more difficult. This paper describes a comprehensive experimental study of bearing spall progression and a physics-based model being developed for bearing prognostics. The model computes the spall growth trajectory and time to failure based on operating conditions, and uses diagnostic feedback to self-adjust and reduce prediction uncertainty. The predictions compare very well to fault progression tests on both subscale bearings and full-scale turbine engine bearings. The experimental data has demonstrated that spall propagation is better behaved than once thought and can be predicted with high confidence. For turbine engine core thrust bearings with a typical mission mix, the prognostic window (first detection to failure) is on the order of 100 flight hours, which provides ample opportunity to plan future missions and maintenance activities with considerable safety margin. Since the model is physics-based, it is readily adaptable to most bearing applications, and a reduced order version has also been developed which is efficient enough to run on-board with little or no loss in accuracy