We argue that the semantic analysis of task-irrelevant stimuli is modulated by feature-specific attention allocation. In line with this hypothesis, we found semantic priming of pronunciation responses to depend upon the extent to which participants focused their attention upon specific semantic stimulus dimensions. In Experiment 1, we examined the impact of feature-specific attention allocation upon affective priming. In Experiment 2, we examined the impact of feature-specific attention allocation upon nonaffective semantic priming. In Experiment 3, affective relatedness and nonaffective semantic relatedness were manipulated orthogonally under conditions that either promoted selective attention for affective stimulus information or selective attention for nonaffective semantic stimulus information. In each of these experiments, significant semantic priming emerged only for stimulus information that was selectively attended to. Implications for the hypothesis that the extraction of word meaning proceeds in an automatic, unconditional fashion are discussed. (Contains 1 table.)