This study puts the Cognition Hypothesis (Robinson 2005) to the test with respect to its predictions of the effects of changes in task complexity ([plus or minus] few elements) and task condition ([plus or minus] monologic) on L2 performance. 44 learners of Dutch performed both a simple and a complex oral task in either a monologic or a dialogic condition. The performance of the L2 learners was analysed with regard to linguistic complexity, accuracy, and fluency. As predicted by the Cognition Hypothesis, the complex task generated more accurate though less fluent speech. Linguistic complexity, however, was only marginally affected. Dialogic tasks triggered more accurate and fluent output though it was structurally less complex. The interaction of task complexity and task condition showed effects on measures of accuracy only: in the monologic but not in the dialogic condition task complexity did promote accuracy. As a consequence, our results only partially support the Cognition Hypothesis. (Contains 7 tables.) [This study was made possible by the Amsterdam Centre of Language and Communication (ACLC) of the University of Amsterdam (UvA).]