This paper presents evidence in support of the idea that observations about surface contrast can inform a language learner about the content of underlying forms. Specifically, it is shown that contrasting surface realizations of a pair of morphemes in a given morphological environment can provide information about the underlying forms of the contrasting morphemes, information about the features on which the surface realizations differ. The authors review formal results that provide sufficient conditions under which definite conclusions about underlying features can be made. A learning procedure for exploiting such contrast information is given, and simulation results of a larger learning algorithm using the contrast learning procedure are presented. An analysis of the behavior of the learning algorithm on different languages reveals interesting trade-offs between the contributions to overall learning made by phonotactic learning and contrast-based learning, depending upon properties of the language being learned.