In this paper, we propose a theoretical framework within which to evaluate the reliability of promises that an agent makes, based on past performance of the agent. Our framework does not just propose one such measure, but defines axioms that govern the choice of measure. The framework is able to account for partial fulfillment of promises, late fulfillment of promises, fulfillment of variants of promises, and the like. Within this framework, we propose some specific measures to evaluate promises made by agents and develop algorithms to compute these efficiently. We tested our methods on a real world data set of airline flight information and show that our methods are both accurate and quickly computable, even on large data sets.