When augmented by suitable auxiliary information, retrospective data can identify response probabilities. The auxiliary information may take the form of data on marginal distributions or appropriate structural assumptions. When a combination of retrospective observation and auxiliary information suffices in principle to identify response probabilities, practical use of this fact requires that statistically sound and computationally tractable estimation methods be available. This article analyzes the problem of identification and presents the needed estimators. © 1976 Taylor & Francis Group, LLC.