Failure to "activate" relevant, existing background knowledge may be a cause of poor reading comprehension. This failure may cause particular problems with inferences that depend heavily on prior knowledge. Conversely, teaching how to use background knowledge in the context of gap-filling inferences could improve reading comprehension in general. This idea was supported in an experimental study comprising 16 sixth-grade classes ("N" = 236) randomly assigned to experimental or control conditions. In the experimental condition, students' contribution to "gap-filling" inferences with expository texts were made explicit by means of graphic models and inference-demanding questions. After eight 30-min sessions, a large training effect was found on students' inference making skills with a substantial and sustained transfer effect to a standard measure of reading comprehension. The effects were not mediated by students' motivation, decoding ability, vocabulary, or nonverbal IQ.