Five experiments investigated the effects of word frequency, neighborhood size, and bigram frequency on lexical decision and word-naming performance. Large neighborhood size, manipulated independently of bigram frequency, facilitated lexical decision and standard naming latencies for low-frequency words but had no effect on delayed naming performance. Bigram frequency, manipulated independently of neighborhood size, had no effect on lexical decision or naming performance. The data suggest that effects of neighborhood size reflect lexical similarity rather than orthographic redundancy and that they are due to lexical access rather than processes specific to lexical decision or naming tasks. The results are incompatible with models assuming that lexical access involves a serial comparison process. The implications for parallel models assuming localized and distributed representations are discussed.