In this paper I investigate the causal returns to education for different educational groups in Germany by employing a new method by Klein and Vella (2010) that bases identification on the presence of conditional heteroskedasticity. Compared to IV methods, key advantages of this approach are unbiased estimates in the absence of instruments and parameter interpretation that is not bounded to local average treatment effects. Using data from the German Socio-Economic Panel Study (SOEP) I find that the causal return to education is 8.5% for the entire sample, 2.3% for graduates from the basic school track and 11% for graduates from a higher school track. Across these groups the endogeneity bias in simple OLS regressions varies significantly. This confirms recent evidence in the literature on Germany. Various robustness checks support the findings.