One of the main drawbacks of the Smith predictor (SP) for controlling systems affected by external delays is that the resulting performance relies on the accuracy of the knowledge regarding such delays. Thus, as the mismatch between the true value of the delay and the nominal one employed for control purposes increases the closed loop performance deteriorates accordingly. This point is particularly remarkable when dealing with multivariable plants in which strong coupling between the different input-output channels may exist. In this communication an intelligent control scheme aimed at the control of multivariable plants with uncertain external delays is proposed. It arises as the result of combining the basic SP with a supervised multiple model architecture that tackles the delay uncertainty.