The purpose of this paper is to present an extension to some of the current work on worst-care identification problems to multivariable systems, We consider an H-proportional to-identification problem for a class of linear shift invariant multi-input/multi-output systems. Out main results are an interpolatory algorithm and a number of bounds on the identification error, This algorithm operates on available input and output data in the time domain and is constructed by solving an extended matrix tangential Caratheodory-Fejer problem. Similar to its counterpart for scalar systems, this interpolatory algorithm possesses certain desirable optimality properties and carl Be obtained via standard convex programming methods.