It was predicted that children learning to read inconsistent orthographies (e.g., English) should show considerable flexibility in making use of spelling–sound correspondences at different unit sizes whereas children learning to read consistent orthographies (e.g., German) should mainly employ small-size grapheme–phoneme strategies. This hypothesis was tested in a cross-language blocking experiment using nonwords that could only be read using small-size grapheme–phoneme correspondences (small-unit nonwords) and phonologically identical nonwords that could be decoded using larger correspondences (large-unit nonwords). These small-unit and large-unit nonwords were either presented mixed together in the same lists or blocked by unit size. It was found that English children, but not German children, showed blocking effects (better performance when items were blocked by nonword type than in mixed lists). This suggests that in mixed lists, English readers have to switch back and forth between small-unit and large-unit processing, resulting in switching costs. These results are interpreted in terms of differences concerning the grain size of the phonological recoding mechanisms developed by English and German children.