This thesis consists of two sections analysing aspects of associative memories. The first section compares the usefulness, limitations, and similarities of the sparse distributed memory (SDM), the cerebella model articulation controller (CMAC) and the Hopfield network. This analysis leads in the second section to a proposal for combining CMAC with a form of robot learning through exploration, the PURR-PUSS system. It is then demonstrated the combination of the PURR-PUSS and CMAC systems produce a system capable of robot control. There are a number of critical factors in the performance of a neural network as a memory. These include the capacity and the efficiency of the training. Of the three networks considered, the Hopfield network is by far the most common in the literature. In spite of this, this thesis shows that the SDM and CMAC are almost identical and, in fact, have significant advantages over the Hopfield network in terms of capacity. This is particularly evident in the storage of sequences, where the SDM shows a significant improvement over the Hopfield network. The major contribution of this thesis is the analysis and development of the full potential of the SDM for data storage. The first contribution is a correction of an error in the existing analysis of the capacity of the SDM. The corrected figure is verified both theoretically and experimentally. The second contribution is an improvement in capacity resulting from an alternative method of generating the outputs. Finally, the capacity is further improved, by using an iterative approach to information storage previously employed on the Hopfield network. The latter approach helps produce a significant advantage in capacity for SDM. Another contribution of this thesis is the combination of associative memory with the a means of learning through experimentation. The PURR-PUSS system was originally developed as a means to enable a robot to learn through interacting with its environment. It is shown that its strengths and weaknesses complement those of the CMAC and SDM systems. PURR-PUSS and CMAC are combined and the result is a system which is capable of superior control than either system by itself This is demonstrated through an example, in which the combined system learns to control a ball rolling in a tilting maze of unknown dynamics. The system begins by learning through random exploration controlled by the PURR-PUSS system. As the knowledge of the environment increases, the PURR-PUSS system is able to successfully achieve goals, although the quality of the control is poor. However the addition of CMAC which in turn learns from PURR-PUSS's movements produces an improvement in the quality of the control.