Four experiments consider the role of semantic category information in word identification using a serial classification reaction time paradigm. Experimental variables were manipulated by varying the semantic properties of blocks of trials and the target search instructions. Experiments 1, 2, and 3 investigated the facilitation on target word recognition of manipulating the categorical homogeneity of the nontarget words. An homogeneous nontarget set facilitated the classification of unrelated target words. This category contrast effect was obtained in all conditions, although its magnitude depended upon target search instructions. Experiment 3 also investigated whether word identification was inhibited if subjects were prevented from utilizing the category information to distinguish target from nontarget words. Target and nontarget word identification was slower under such conditions. Experiment 4 demonstratd that both the category contrast and category interference effects were dependent upon the use of short response-stimulus intervals to define a functional semantic background. This suggests the category effects are perceptual in nature. Current models of word recognition cannot easily explain the findings. A committee decision model is outlined to accommodate the data; the model proposes that visual analysis, identification, and categorization proceed in parallel.