We are concerned with assuring the reliability of semiconductor lasers intended for an application in which the design lifetime is long, replacement or redundancy is impossible or impractical, and the failure of even a few lasers could be disastrous. In the search for a reliability assurance strategy that will meet our objectives, we have carefully examined the well-known and widely used bathtub and lognormal approaches. Based upon our understanding of the expected aging behavior of lasers, we propose an alternative reliability assurance strategy that we believe to be an improvement over the traditional approaches. The object is not how to make reliable lasers, but rather how to confidently predict, in a timely fashion, which lasers in a given population will endure beyond the intended system lifetime. Particular emphasis is placed upon initially imposed overstress regimes that address the anticipated presence of transient modes of depadation and infant failure modes with low thermal activation energies that may be invulnerable to detection during accelerated thermal aging. Since lasers degrade gradually rather than fail suddenly, comparable emphasis is placed upon monitoring the stabilized long-term degradation rates of the survivors of the overstress regimes so as to permit lifetime predictions of individual lasers.