Very often, the control of multivariable industrial processes is performed using multi-loop architectures, in which several SISO PI or PID controllers are committed to control different channels of the plant. A difficulty with such a strategy arises due to the interaction between the different control loops, which may cause the control action in a loop to give rise to significant disturbances in other loops. This paper presents a new robust PI tuning procedure for multi-loop control systems which aims to decouple the different channels of the multivariable system and to guarantee the tracking response performance and measurement noise attenuation, using a reference model approximation scheme. The control problem is formulated as a non-convex multi-objective optimization problem which is formulated directly in the space of PI controller parameters. Polytopic models represent the system uncertainty. An application example is developed for the control of a quadruple-tank process.