IMC controller design for a process is based on choosing a mathematical model that describes the real process. The mathematical model describing such process is often not unique because the real variables of the process can vary within an interval. In such cases the performance of the control system varies, possibly substantially, as process parameters change. To limit such performance variations, we have developed an algorithm for choosing the model gains and the filter time constants of the IMC controller, to minimize the amount of interaction between outputs due to set point changes and disturbances for multivariable decoupled control systems. Some examples illustrate the algorithm.