This study investigates whether practice with computer-administered feedback in the absence of meaning-focused interaction can help second language learners notice the corrective intent of recasts and develop linguistic accuracy. A group of 218 beginning Anglophone learners of Spanish received 1 of 4 types of automated feedback (no feedback, utterance rejection, recasts, or enhanced recasts) in the written mode or the oral mode, in response to noun--adjective gender or number agreement errors. For both modalities, written and oral post-tests conducted up to 2 months after treatment revealed that recasts yielded more target-like production and learner repair than either no feedback or utterance rejection and that utterance rejection was in turn more effective than no feedback. Two factors were found to increase the effectiveness of computer-delivered recasts: oral, but not typographical, input enhancement (orally enhanced recasts were more beneficial than orally unenhanced and typographically enhanced recasts) and working memory (higher span learners were superior to lower span learners in all recast groups). We discuss the implications of our results for the area of error treatment as an instructional focus technique and suggest avenues for future research. (Contains 4 tables, 3 figures, and 4 notes.)