Retrieval practice promotes learning. In typical retrieval practice, learners first encountering a new study item are given the expected answer along with the cue (an initial passive study trial). Withholding the answer by giving only the cue (an initial attempted retrieval trial) creates an additional retrieval opportunity. In contrast to earlier lab studies, we examine whether initial attempted retrieval promotes learning in realistic situations, where (a) learners can be expected to have varying degrees of prior knowledge of the study materials; (b) items are repeated multiple times; and (c) learners receive continuous feedback. In four experiments (N = 210), we contrast conditions with initial attempted retrieval to initial passive study in a multiple-repetition retrieval practice session. In the first two experiments, we demonstrate that initial attempted retrieval results in higher retrieval accuracy and faster response latencies on subsequent repetitions, regardless of the success of the initial retrieval attempt. However, after multiple repetitions of an item, benefits of initial attempted retrieval diminish, calling into question their practical usefulness. In the remaining two experiments, we show that initial attempted retrieval is beneficial in multiple-repetition learning when the attempted retrieval trials are used to identify prior-knowledge levels in adaptive learning applications. Dropping items that were answered correctly on the first attempt enhanced overall learning efficiency, especially for learners with moderate to high prior knowledge, but without any disadvantage for learners with low prior knowledge. These results demonstrate under which conditions the attempted retrieval benefit can improve learning outcomes in the classroom.