The "memory" literature emphasizes the notion of encoding variability, whereas the "word recognition and attention" literature typically asserts that it is impossible to prevent immediate lexical and semantic activation when single words are presented to skilled readers. In the four experiments reported here, the presence/absence of a Stroop effect was associated with the nature of the between-trial context. These data can be understood as an expression of contextually driven encoding variability, but are problematic for the decontextualized "automatic" processing account that has prevailed in the attention and performance literature for the last six decades.