Teaching previously learned contents to (fictitious) peers is regarded as a beneficial activity that aids learning. However, it is still an open question which cognitive mechanism (generating versus retrieving information) account for this learning-by-teaching effect. To examine the role of retrieval during generative processing while learning-by-teaching, we conducted an experiment. After a learning phase about the human respiratory system, university students (N = 108) either taught the learned contents with (open-book teaching) or without (closed-book teaching) access to the learning material. Students in the control group restudied the contents. Results indicated that teaching was not more beneficial than restudying regarding students' learning outcome (i.e., retention, drawing knowledge, transfer knowledge). However, open-book teaching was more effective than closed-book teaching regarding students' retention. This effect was mediated by the richness (i.e., number of concepts) and completeness of students' explanations. The findings suggest that retrieval failure may constrain potential effects of closed-book explaining.