It is not realistic to calculate failure rates by building many units and running them for many hours, under expected operating conditions. This is especially true for well-designed and properly built supplies, with low failure rates. Instead two key methods can be used, prediction during design, and assessment during manufacturing. A third class, observation is also possible but this is more difficult as it is impossible to control all of the conditions a supply has been subjected to and therefore more difficult to undertake reliable causation analysis. Prediction uses one of several standard databases of component failure rates and expected lives, among them are MIL-HDBK-217 (U.S. Navy), HRD5 (British Telecom), and Telcordia. It is important to be consistent in the prediction methodology and database, and in general, MIL-HDBK-217 is focused on military and commercial applications, while the Telcordia document concentrates on telecommunications-oriented designs and applications. The MIL approach requires use of many parameters for the different components and includes voltage and power stresses, while Telcordia requires fewer component parameters and can also take into account lab-test results, burn-in data, and field-test data. Finally, the MIL approach yields MBTF data, while Telcordia produces FIT data, failures per billion hours. However, use of these databases and techniques does not guarantee absolute accuracy, as it is based on assumptions which are somewhat incorrect, at best. It assumes that the design is perfect, the stresses are all known, everything is operated within its ratings, any single failure will cause complete failure, and the database is current and valid. The reality is that often databases are quite old and don’t have data on newer components.