Traffic signal control is very important to urban development. Most of the governments in developed countries continue to improve the safety as well as efficiency of their traffic signal control systems. In a well-designed traffic control system, fewer traffic accidents will occur. Moreover, the vehicle waiting times can also be shortened. In this thesis, we attempt to apply adaptive learning techniques to traffic signal control. In particular, emphasis is put on three important aspects: (a) Adaptation: The controller should adapt its signal time plan to varying traffic patterns that are characteristic of non-stationary stochastic environments. (b) Memorization: The controller should be capable of remembering previously encountered traffic patterns so that there is no need to re-learn the patterns when these situations are encountered again. (c) Cooperation: Controllers for different traffic junctions should interact and cooperate with each other to optimize global performance. Several control strategies are studied in the thesis. First, the regular time method is a non-adaptive method that serves as a baseline for comparison with other methods. An adaptive extension of the regular time method, called the adaptive regular time method, improves performance through its adaptiveness to traffic changes. This adaptive method is then further generalised by formulating the adaptive traffic signal control problem as a reinforcement learning problem. Two reinforcement learning methods, called LRP and Q-learning respectively, are studied. It is found that reinforcement learning is very effective for implementing adaptive traffic signal control systems that can satisfy the three criteria above.