Self-Tuning control is an attractive solution to the problem of controlling unknown, time-varying and noisy processes, where conventional PXD controllers may give unsatisfactory performance.In this research work, an attempt has been made to bring the Self-Tuning control methods from the academic world towards industrial practice, and to produce a 'Self-Tuner* hardware, with a performance superior to that of PID controllers, at a comparable cost. Overall, the above objective was achieved in this work.The thesis investigates Self-Tuning and the related theory, and examines the hardware/software requirements of a microprocessor-based Self-Tuner. The principles and characteristics of sampled-data systems with regard to their effects on Self-Tuning are discussed. Various recursive-identification techniques are investigated. It is shown that a Square-Root algorithm based on Least-Squares (LS) identification technique is suitable for implementation on an 8 bit microprocessor. Various methods to track the changing parameters of a process are discussed and a scheme for employing a dynamic forgetting-factor in an LS estimation algorithm is presented.Current Self-Tuning techniques are described. From a study of Self-Tuning and Model Reference Adaptive Control schemes, it is demonstrated that the two approaches can be unified. A new Self- Tuning strategy, herein termed Model-Referenced Minimum Variance (MRMV) Self-Tuner is proposed as a general case of Minimum Variance Self-Tuners. It is shown that this new scheme results in softer control and more acceptable practical performance.A simulation package for studying various Minimum-Variance Self-Tuners is presented. Simulation results of Self-Tuning control over various plants dynamics are given. The simulation results verify the advantages of the proposed MRMV control scheme.The hardware aspects of a Self-Tuner, based on the Z80 microprocessor are described. Two methods of performing complex arithmetic, using software routines and a 'number-cruncher* chip are presented and compared. Due to the self-compensating nature of the algorithms, it is shown that undemanding analogue interfacing requirements only are necessary. The assemble language programs to use the described hardware and implement the discussed Minimum-Variance controllers are presented. The memory requirements and the execution speed of the routines are also given. The concept of 'mathematical noise injection' is introduced as a cure for possible ill-conditioning problems associated with the covariance matrix in the etimation routine.Various experiments are performed with the implemented Self- Tuner to control different analogue processes and the results are given. It is shown that the 'tuning' of the Self-Tuners (limited to the choice of 3 or 4 parameters) is not critical and good performance may be obtained with a wide choice of the tuning parameters.The application of Self-Tuners to the ship-steering problem is discussed. Several experiments on board two similar cargo ships are performed and the results suggest that Self-Tuners could be used for this purpose with significant advantage.