Theories of generative linguistics hold that language processing occurs by means of the manipulation of symbols by explicit rules. The past 15 years have seen the development of a radical challenge to this view, derived from a form of computational modelling called Parallel Distributed Processing or connectionism. Connectionist linguistic theory holds that language processing takes place at a subsymbolic level, and that the appearance of rule-driven behaviour is formed by the abstraction of patterns from the environment. The English past tense has become a critical arena of dissent between symbolic and subsymbolic theories of linguistics. There are two current dominant theories of past tense inflection: hybrid dual-route theory, championed by Pinker, posits a symbolic, explicit, default rule for the regularisation process, and an associative memory component for irregular exceptions; single-route theory maintains that regular and irregular inflectional morphology may both be accounted for within a single, subsymbolic, associative system which contains no explicit linguistic rules. This doctoral thesis describes a new classification of phonological similarity between verbs ('neighbourhood density'), which is used to develop mutually-exclusive and empirically-testable hypotheses from the two dominant theoretical perspectives of English past tense inflectional morphology. Empirical research is conducted in the domains of experimental psychology, electrophysiology and Connectionist modelling: four novel verb elicitation tasks are performed on adults; two ERP studies investigate brain activity for regular, irregular and novel verb inflection; and five neural network simulations are built in order to compare human data with network performance. Data are reported which have implications for single- and dual-route theories of past tense processing.