This article reviews and compares three cognitive process models that relate hindsight bias to changes in an underlying knowledge base as a result of outcome feedback. Two of these models, SARA (Selective Activation and Reconstructive Anchoring; Pohl, Eisenhauer, & Hardt, 2003) and RAFT (Reconstruction After Feedback with Take the Best; Hoffrage, Hertwig, & Gigerenzer, 2000), focus on memory processes involved in hindsight bias and have been implemented as computational models. The third, focusing on processes affecting hindsight probability judgments, is a new synthesis and elaboration of the influential work of several authors (e.g., Hawkins & Hastie, 1990; Pezzo, 2003; Roese & Olson, 1996), which we present here in a more systematic way under the name of causal model theory. We compare the three models with respect to their ranges of applicability, the types of postulated changes in the knowledge base, and the nature of the featured cognitive processes (e.g., unconscious/automatic versus conscious/flexible). Finally, we outline ways to combine the main processes featured in the models within an integrative theoretical framework and suggest testable hypotheses for future research.