In some situations, rumors come to reflect reality quite well and in others they become fantastic. Why? This article summarizes research on rumor accuracy. It considers rumors as unverified statements in circulation, and rumor discussion as serving group sensemaking and threat-management functions. It also explores a complementary view of rumors as memes: adaptive statements that survive or die out. Conceptualizations of rumor accuracy, methods of investigation, prospects for future social psychological study, and accuracy rates are examined. The Percolation Model of rumor accuracy is reviewed: rumor distortions and errors that occur in part because of cognitive limitations and perceptual biases are either compounded by relationship- or self-oriented aims and a circumscribed ability to test the rumor, or countervailed by accuracy motivation, message checking, and veracity checking; rumor activity hastens these processes. The multivariate, interactive, and dynamical systems nature of rumor accuracy phenomena is highlighted. The implications of this research for social epistemics are discussed.