The present study investigated strategic variation in reliance on phon
ological mediation in visual word recognition. In Experiment 1, semant
ically related or unrelated word primes preceded word, pseudohomophone
(e.g., trane), or nonpseudohomophone (e.g., trank) targets in a lexic
al decision task. Semantic priming effects were found for words, and r
esponse latencies to pseudohomophones were longer in related than in u
nrelated prime conditions. In Experiment 2, related or unrelated word
primes preceded word or pseudohomophone targets. A relatedness effect
was found for words, although it was significant at a 600-msec prime-t
arget stimulus onset asynchrony (SOA) and not at a 200-msec SOA. There
was no relatedness effect for pseudohomophones. Experiment 3 was a re
plication of Experiment 2, except that pseudohomophones were replaced
by nonpseudohomophonic orthographic controls. Facilitation effects for
related target words were greater in Experiment 3 than in Experiment
2. The results reflect apparent variations in the expectation that a r
elated prime reliably indicates that a target is a word. Although reli
ance on phonological mediation might be strategically contingent, ther
e could be a brief time period in which phonologically mediated lexica
l access occurs automatically. Whether phonological information is mai
ntained or suppressed subsequently depends on its overall usefulness f
or the task.