The way in which the behavioral manifestations of autism spectrum disorder (ASD) emerge in infancy is variable. Regression-loss of previously acquired skills-occurs in a subset of children. However, the etiology and significance of regression remains unclear. Until recently, investigation of regression relied on retrospective report by parents or examination of home videos from early in life. However, home videos and retrospective report of the nature and timing of regression, and association with factors such as illness or immunization, is potentially subject to bias. The advent of prospective studies of infant siblings at familial high-risk of ASD has the potential to document regression as it occurs. Recent research has suggested that subtle loss of skills occurs in a larger proportion of children with ASD than previously assumed; however, there are few reports of clear-cut regressions, such as that involving dramatic loss of language and other established skills, in the prospective literature. This could be because of the following: clear-cut regression occurs less commonly than parent report suggests, study design limits the potential to detect regression, or there are differences between multiplex and simplex families in the rate of de novo genetic mutations and therefore regression risk. This review will bring together literature from retrospective and prospective research and attempt to reconcile diverging findings, with a specific focus on methodological issues. Changing conceptualizations of regression will be discussed, as well as etiological factors that may be associated with regression. The main challenges that need to be addressed to measure regression in prospective studies will be set out.Lay Summary: Regression-a loss of previously established skills-occurs in a subset of children with ASD. Parental recall is not always accurate but studying younger siblings of children with ASD, 10-20% of whom will develop ASD, should make it possible to measure regression as it occurs. Clear-cut regression, like loss of language, has not often been reported in infant sibling studies, but recent research suggests that gradual loss of social engagement might be more common. This review looks at the evidence for regression from infant sibling studies and asks how study design affects the likelihood of capturing regression.