The goal of this study is to investigate the impact of manipulating the cognitive complexity of three different types of oral tasks on interaction. The study first considers the concepts of task complexity and interaction and then examines the specific studies that have looked at the effects of increasing task complexity on conversational interaction. In the experiment, learners of English as a foreign language organized into 27 dyads carry out three different types of tasks: a narrative reconstruction task, an instruction-giving map task, and a decision-making task. Two different versions of each task, one simple and one complex, are presented to learners in different sequences. Task complexity is manipulated along the degree of displaced, past time reference, the number of elements, and the reasoning demands. Audio recordings are transcribed and coded for interactional feedback, which is measured in terms of negotiation of meaning (i.e., confirmation checks, clarification requests, and comprehension checks), recasts, language-related episodes (LREs), and repairs, all of which have been described in the literature as being conducive to acquisition. Both parametric and non-parametric statistical tests are used. Results are discussed in the light of previous studies that have looked at the specific relationship between task complexity and interaction, attention models (Robinson, "Applied Linguistics" 22: 27-57, 2001a, "Second Language Studies" 21: 45-105, 2003, "International Review of Applied Linguistics" 55: 1-32, 2005, "International Review of Applied Linguistics" 45: 193-213, 2007b; Skehan and Foster, "Cognition and tasks," Cambridge University Press, 2001), and how different task types may variously affect the way interaction proceeds during task performance.