This is an industrial case study on the introduction of self-tuning control into a selected set of control loops in a large oil refinery. A system identification methodology was used to determine which process loop is undergoing change in its dynamics, or that the loop disturbances are changing. A statistical cluster analysis technique was used to select data duration where the process did not change its operating condition. A multi-input/single-output system identification technique based on the prediction error method and ARMAX models was used to determine the loop model for a given operating region. To establish whether a given loop/process is undergoing any major change in gain, dominant time constant, or time delay, cross validation of models between operating conditions was used. This proved that changes were taking place in temperature loops of two high temperature steam reformers. The parameter identification included both plant and disturbance models. Justification for using self-tuning was based on analysis of process parameter fluctuation and corresponding controller parameter changes. Actual field results on production scale steam reformers with self-tuning controllers are presented.