Substantial efforts have been spent on characterizing the round complexity of various cryptographic tasks. In this work we study the round complexity of secure multiparty computation in the presence of an active (Byzantine) adversary, assuming the availability of secure point-to-point channels and a broadcast primitive. It was recently shown that in this setting three rounds are sufficient for arbitrary secure computation tasks, with a linear security threshold, and two rounds are sufficient for certain nontrivial tasks. This leaves open the question whether every function can be securely computed in two rounds.We show that the answer to this question is "no": even some very simple functions do not admit secure 2-round protocols (independently of their communication and time complexity) and thus 3 is the exact round complexity of general secure multiparty computation. Yet, we also present some positive results by identifying a useful class of functions which can be securely computed in two rounds. Our results apply both to the information-theoretic and to the computational notions of security.