The huge difference in the level and variance of student performance in the 2000 PISA study between Finland and Germany motivates this paper. It analyses why Finnish students performed so much better by estimating educational production functions for both countries. The difference in the reading proficiency scores is assigned to different effects, using Oaxaca-Blinder and Juhn-Murphy-Pierce decomposition techniques. The analysis shows that German students have on average a more favorable background except for the lowest deciles, but experience much lower returns to these background characteristics in terms of test scores than Finnish students. The results imply that early streaming in Germany penalizes students in lower school types and leads to a greater inequality of educational achievement. It remains unclear, however, if this can be attributed to the effect of school types per se or student background and innate ability that determine the allocation process of students into school types. Overall, the variation in test scores can be explained much better by the observable characteristics in Germany than in Finland.