The study investigated the effects of recasts and prompts on learning language forms that arose incidentally in dyadic interaction, focusing on the degree of explicitness of each type of feedback and learner proficiency levels. The data were collected from 64 beginning and upper-intermediate English learners of Korean. Thirty-one beginning learners were randomly assigned to a feedback group and to a control group, and thirty-three upper-intermediate learners were also assigned to a feedback group and to a control group. Each learner was paired with an English speaker and worked on a picture-sequencing task. The learners in the feedback groups received recasts or prompts on their erroneous utterances while the learners in the control groups did not. The effects of feedback were measured by pre-interaction picture descriptions and immediate and delayed post-interaction correction tasks. The study found that both recasts and prompts had some effects on learning the targeted forms, and more explicit forms of each feedback resulted in a higher rate of correction. The beginning learners took more advantages of recasts relative to the upper-intermediate learners, and explicit prompts worked better for the upper-intermediate learners. The effects of prompts sustained longer than recasts in both levels.