The problem of robust linear transceiver design in Multiple-Input Multiple-Output (MIMO) ad hoc Cognitive Radio Networks (CR-Nets) is studied in this paper. In this problem multiple interfering MIMO links are active within the service range of a Primary Radio Network (PR-Net) for which, unlike in conventional design problems, the Channel State Information (CSI) is not known perfectly. The imperfection in CSI is modeled using norm-bounded uncertainty matrices. The design problem is formulated to provide the minimum Sum-Mean Square Error (SMSE) of all the links while the transmit power of Secondary Users (SUs) is limited and the amount of interfering power toward the Primary users (PUs) is controlled. This problem is not convex, because its objective function is non-convex in nature and it has semi-infinite robust constraints. To overcome these limitations an iterative solution, which is based on the relaxed version of this problem, is provided. The problem is then solved numerically. Finally simulation results are provided to demonstrate the performance of this method.