This paper employs the dynamic treatment effects methodology proposed by Heckman et al. (2016, 2017) to examine educational transitions and expected returns in the German education system which is characterized by rigid early tracking but with options to revise track choices at later stages. We document strong sorting of individuals along observed and unobserved characteristics across the stages of the system. We consider expected wage returns to track choices including the continuation values arising from the options opened up by choosing a certain track. Expected returns to choosing higher tracks are generally positive but highly heterogenous. We find sorting on gains at many but not all stages of the system. A considerable percentage of the population exercises 'second chance' options to revise earlier track choices. The value of these options strongly depends on parental background as individuals from higher backgrounds are better able to exploit the possibilities opened up by these options at later stages. We present estimates of wage returns to different forms of vocational and academic training free of ability and sorting bias. Returns to academic training are particularly heterogenous.