Informed by the cognition hypothesis (Robinson, 2011), recent studies indicate that more cognitively complex tasks can result in better incorporation of feedback during interaction and, as a consequence, more learning. It is not known, however, how task complexity and feedback work together in computerized environments. The present study addressed this gap by investigating how cognitive complexity in face-to-face (FTF) versus computer-mediated communication (CMC) environments mediates the efficacy of recasts in promoting second language development. Eighty-four adult learners of Spanish as a foreign language at a mid-Atlantic university were randomly assigned to a control group or one of four experimental groups. The experimental groups engaged in one-on-one interaction and received recasts on the Spanish past subjunctive but differed according to (a) whether or not they had to reflect on another person's intentional reasons during the task and (b) whether they interacted in FTF or CMC environments. Learning was measured with two production tasks and a multiple-choice receptive test in a Pretest-Posttest 1-Posttest 2 design. Results revealed that in the FTF mode, performing the cognitively complex task while receiving recasts led to the most learning. In the CMC mode, the cognitively complex task + recasts was not effective. Instead, the cognitively simple task led to the most development in CMC. The study also found that judgments of time on task were the only independent measure of cognitive complexity that held across mode.