Some of the common methods of reliability prediction utilizing calendar or lot size dependence and various "k" factors for complexity, end use, product family, and environment are briefly reviewed. These techniques, although sometimes accurate and simple, do not provide adequate reliability tradeoff information and do not fully treat the effects of vendor variability and improvements in the state-of-the-art. From these techniques it is frequently difficult to understand causes of observed failure rates and to determine what can be done to achieve cost-effective reliability.One method of alleviating these objections is to derive a model based on microcircuit failure mode and mechanism knowledge. The current status of microcircuit failure knowledge is reviewed. Considerable use is made of data which has been acquired at very high stress levels and the relationship between this data and end use data is discussed.Microcircuit failures can be grouped into four essentially independent categories: failures due to time degradation, failures due to mechanical stress, serial failures due to two or more events occurring in succession, and parts which never worked. A failure prediction model based on failure modes and mechanism knowledge would contain each of these elements. The resulting model might prove unwieldy and of dubious value because of its size. However, a simplified model can be employed. If the user knows how it was derived, it can be a useful and flexible tool. One form that such a model might take is presented and discussed.