This study examines the role of task-based conversation in second language (L2) grammatical development, focusing on the short-term effects of both negative feedback and positive evidence on the acquisition of two Japanese structures. The data are drawn from 55 L2 learners of Japanese at a beginning level of proficiency in an Australian tertiary institution. Five different types of interactional moves made by native speaker interlocutors during task-based interaction were identified, by way of which learners received implicit negative feedback and positive evidence about the two target structures. The relative frequency of each interactional move type was calculated, and associated changes in the learners' performance on immediate and delayed posttests were examined. It was found that, although native speaker interactional moves containing positive evidence about the two target structures were 10 times more frequent during task-based language learning than those containing implicit negative feedback, only learners who had an above-average score on the pretest benefited from the positive evidence provided. Implicit negative feedback, on the other hand, had beneficial effects on short-term development of the grammatical targets regardless of the learner's current mastery of the target structures. Moreover, recasts were found to have a larger impact than other conversational moves on short-term L2 grammatical development.