I begin by reviewing the sentence processing field and assessing the legacy of generative grammar and its associated view of the mind on proposed mechanisms of language processing. I argue that this legacy prevents many models of processing from being able to account, in principle, for recent findings within the field. A model of processing is proposed which combined insights from Cognitive Grammar and constraint-based parsing, including the principle of parsing as categorisation proposed by Langacker (2000). This model, the categorising, constraint-based parser, is tested alongside parsing models from the literature in the remainder of the thesis. A series of ten experiments is presented which look at the effect of “head distance” on temporarily ambiguous sentences of English. Head distance refers to the distance between the lexical head of an initially ambiguous part of the sentence (“soldier” in the examples below), and the disambiguating region (“reached…”). This distance is greater in (1b) than in (1a). (1a) as the young woman was waking the young and bearded soldier reached for his sword. (1b) as the young woman was waking the soldier who was young and bearded reached for his sword. Distant conditions have been found to be judged less grammatical than Close conditions (Ferreira and Henderson, 1991b). However, until very recently theoretical accounts of reanalysis incorporating this finding had been based on off-line measures rather than real-time processing. Head distance appeared to have no effect on real-time processing and to be limited to one ambiguity type. The categorising, constraint-based parser developed in this thesis provides an explanation for this effect, and predicts the types of environment in which it should occur. Investigation into this effect can also shed further light on the nature of the reanalysis process, providing data to test explanations derived from commitment- and repair-based accounts of reanalysis; and, for activation-based models, to question whether the notion of decay and re-activation is sufficient to account for processing costs. A self-paced reading methodology is used to assess the impact of head distance on reading times for a variety of ambiguity types. Grammatically judgement and comprehension data is also gathered.