The present quasi-experimental study investigated the comparative effects of metalinguistic clue (MC) and metalinguistic explanation (ME) on the accurate use of Japanese transaction expressions. The study recruited 25 learners in a second-semester Japanese course (i.e. non-introduced group) and 17 students in a fourth-semester course (i.e. introduced group) at an American university. They worked on computer-delivered picture description practice with either type of feedback. Their production during the practice was analyzed in terms of successful uptake, and the test instrument consisted of timed oral production and untimed true/false judgment. The results from the uptake analysis indicated that the ME elicited significantly more successful uptake than the MC. However, multiple regression analyses showed that successful uptake was only marginally predictive of immediate performance on the timed oral production. In addition, a pretest, immediate posttest, and delayed posttest demonstrated that the ME led to retention for the introduced group, while it did not for the non-introduced group. Meanwhile, the effects of the MC were not moderated by prior knowledge. The results were attributed to the different cognitive processing of feedback engendered from feedback type and prior knowledge.