The problem of minimizing the sensitivity of the output or input of a plant under a feedback control with respect to external additive disturbances is considered. The mathematical model assumed for the plant is a rectangular matrix of transfer functions which can arise from both lumped-parameter and distributed-parameter systems (including time delays). The feedback controller is assumed to guarantee the closed-loop stability. It is shown that an arbitrarily small sensitivity function on a compact set of complex frequencies can be obtained by a suitable choice of feedback transfer function, provided no zero of the plant transfer function is in this set. Additionally, the sensitivity function can be designed to be bounded on an unbounded region containing the imaginary axis, if there are no zeros of the plant in the right half-plane and a certain hypothesis holds. Multiplicative delays are not excluded. If there are zeros in the right half-plane the latter design is not realizable, since minimizing the sensitivity on a bounded interval implies its increasing without bounds on the remaining part of the imaginary axis (an appropriate inequality is derived) The paper extends recent results of Zames and coworkers.