This paper investigated if prior context was used by readers to resolve temporarily ambiguous prepositional phrases. The paper addresses two issues, referring to two competing approaches to sentence processing and testing the predictability of each approach in two online reading experiments. The issues focus on the Human Sentence Processing Mechanism (HSPM), looking at whether it is a single structural analysis or multiple structural analyses at points of syntactic ambiguity and how these structural analyses are generated during sentence processing. Two experiments test the predictions of the Garden Path models (the original Garden Path model and the Lexical Filtering Proposal) and the Constraint Satisfaction Approach, concerning the role of sentence context in resolving syntactic attachment ambiguities. Experiment 1 involved 42 undergraduate students, fluent in American English, and 18 experiment items; sentences were presented to subjects on a computer screen. Results did not support the strong lexical guidance predictions of the Constraint Satisfaction Approach and other lexically-driven models. Experiment 2 involved 24 undergraduate students, fluent in American English and using the same experiment items as in experiment 1, except that eye movements were recorded. Results indicate that eye gaze duration and total time data revealed very different patterns. Overall, results are consistent with the Lexical Filtering Proposal of sentence processing, which assumes that initial parsing decisions can be made without reference to nonsyntactic information. (Contains 26 references.) (NAV)