Serial order is central to much of human behaviour including short-term memory. There exists a wealth of empirical data and a number of attempts have been made at providing a theoretical account of these data. However, no existing model accounts for more than a limited subset of the existing data, and no existing model allows examination of developmental improvement at the same time as covering a wide range of adult data. In the following thesis, two models of short-term memory are presented. The first, a developmental associative recall network, DARNET, uses gradient descent based learning to learn how to perform single trial learning and recall of novel paired associates. The second, a neurobiologically plausible oscillator-based associative recall
model, OSCAR, uses Hebbian association to associate items with different states of a reinstatable dynamic control signal, the context. OSCAR is fitted to a range of empirical data including serial position curves, memory span, phonemic similarity effects and item versus order error distributions. Furthermore, it is suggested that if OSCAR is coupled with DARNET, they could provide a developmental account of short-term memory for serial order.