Background: The social gradient in health is one of the most reliable findings in public health research. The two competing hypotheses that try to explain this gradient are known as the social causation and the health selection hypothesis. There is currently no synthesis of the results of studies that test both hypotheses. Methods: We provide a systematic review of the literature that has addressed both the health selection and social causation hypotheses between 1994 and 2013 using seven databases following PRISMA rules. Results: The search strategy resulted in 2952 studies, of which, we included 34 in the review. The synthesis of these studies suggests that there is no general preference for either of the hypotheses (12 studies for social causation, 10 for health selection). However, both a narrative synthesis as well as meta-regression results show that studies using indicators for socio-economic status (SES) that are closely related to the labor market find equal support for health selection and social causation, whereas indicators of SES like education and income yield results that are in favor of the social causation hypothesis. High standards in statistical modeling were associated with more support for health selection. Conclusions: The review highlights the fact that the causal mechanisms behind health inequalities are dependent on whether or not the dimension being analyzed closely reflects labor market success. Additionally, further research should strive to improve the statistical modeling of causality, as this might influence the conclusions drawn regarding the relative importance of health selection and social causation.