The robust control of the crude oil outlet temperature uniformity in a heating furnace of a petroleum refinery is addressed. A reliable dynamic model of the nominal process has been attained using a system identification procedure based on real-time data. This procedure yields a second order model with a dominant time-delay. A PI controller embedded in a modified Smith predictor structure is therefore proposed. Sensitivity and robustness properties of this control system are analytically obtained. Based on that, a tuning procedure is developed for this control system which has lower sensitivity to disturbances than the standard Smith predictor while guaranteeing the system stability when plant parameters change. Simulations are carried out of the proposed control system and other well-known linear advanced process control systems. The comparison of the obtained results shows the superior performance of our control system in most cases, both in rejecting a variety of disturbances and in maintaining the closed-loop stability when the process parameters change.