This work presents a framework to address the MIMO (multiple-input multiple-output) control design problem under model uncertainty. The plant is considered to be LTI (linear and time-invariant) and that it is parametrized by a random variable whose distribution may be unknown but can be sampled from. The objective is to design a dynamic LTI MIMO controller optimizing the closed-loop H2-performance. To account for the uncertainty in the plant, we employ measures of risk on the performance and discuss how coherent measures of risk are suitable to address this problem. In particular, we introduce a measure called Conditional Value-at-Risk (CVaR) and analyze its advantages and disadvantages regarding other traditional measures of risk. A simulation study complements the discussion.