Previous research has demonstrated that increases in probabilistic phonotactics facilitate spoken word processing (e.g., Pitt & Samuel, 1995; Vitevitch & Luce, 1999), whereas increased competition among lexical representations is typically associated with slower and less accurate recognition (e.g., Luce & Pisoni, 1998). We examined the combined effects of probabilistic phonotactics and lexical competition by generating words and nonwords that varied orthogonally on phonotactics and similarity neighborhood density. As predicted, the results revealed simultaneous facilitative effects of phonotactics and inhibitory effects of lexical competition. However, one anomalous result emerged: Certain stimuli with high probability phonotactics and low neighborhood density were processed far more slowly than predicted by a simple model in which effects of phonotactics and density are additive. Using a speeded task in which we asked participants to produce the nearest sound-based neighbor of a spoken target item, we tested the hypothesis that items with high sublexical frequencies but few lexical neighbors have strong lexical attractors that may control processing times under certain specific circumstances. Our results are not consistent with models of spoken word recognition in which degree of lexical competition is a simple function of the weighted sum of the activations of lexical competitors, nor are our results consistent with a model in which a single, highest-frequency neighbor is responsible for all competition effects. Instead, lexical competition seems to be a complex function of the weighted activations of the lexical and sublexical representations associated with the target and its competitors.