Readers' gaze durations are longer on biased homographs (those having one highly dominant meaning) than on unambiguous control words when the preceding context instantiates the homograph's subordinate meaning. In two experiments, we attempted to eliminate this subordinate bias effect by increasing the contextual bias in favor of the homograph's subordinate meaning. In both experiments, subjects' eye movements were recorded as they read target sentences containing biased homographs in which immediate prior context instantiated the subordinate meaning of the word. In the first experiment, subjects were familiarized with the subordinate sense of the critical homograph via a paired-associate task prior to reading the target sentence. In the second experiment, the target sentence formed the last line of an extended passage whose global discourse biased the subordinate meaning of the critical homograph. In addition, the presence or absence of a prior instance of the critical homograph (instantiated in its subordinate sense) at an earlier location in the passage was manipulated. In neither experiment did the increase in contextual bias result in a reduction of the subordinate bias effect. We argue that these results are inconsistent with a selective account of lexical access. (C) 1994 Academic Press, Inc.