Historically, two onset patterns in autism spectrum disorder (ASD) were described: early onset of symptoms and regression in which one-third appear to show a loss of previously established skills in the second year of life. Since this phenomenon could represent a distinct ASD subtype and provide more insight into the etiology, diagnosis, and prognosis, many studies have compared these two groups. The present review discusses definitions, etiology, and methods used in research with a retrospective design and provides an overview of the results on early development and outcomes. However, retrospective research has not provided clear answers on regression as a distinct subtype of ASD and the historic division between early onset and regression does not seem to fit the empirical findings. Based on inconsistent results, future research on onset patterns in ASD needs to be more systematic on the definitions and methods used. Several recommendations to enhance the reliability of future retrospective results are discussed. The combination of a categorical and dimensional approach provides a new interesting framework.