Rapid thermal processing is a new technique in microelectronics technology. Its main characteristics are single wafer processing and fast thermal kinetics (typically 200°C s −1 ). These characteristics imply stringent specifications for temperature control. This paper reports on the study of adaptive temperature control for such a process. The adaptive scheme is obtained by combining a parameter estimator and a long-range predictive controller. Using generalized predictive control and double model reference control, we present and compare the performances of these schemes in real-life experiments, and under different conditions. It is shown that adaptive predictive control performs as well as a PID controller with parameter adjustment, but with a much lower tuning cost.