English-Spanish bilinguals named visually presented words aloud in each language. The words included cognates (e.g., fruit-fruta) and non-cognate translations, (e.g., pencil-lápiz). The cognates were selected so that the orthographic and phonological similarity of their lexical form in each language varied orthogonally. Cognate naming latencies were influenced by the cross-language match of the orthographic and phonological codes. When the orthographic forms were similar in the two languages, naming latencies were slowed by dissimilar phonology, providing evidence for feed-forward activation from orthography to phonology across languages. When the orthographic forms were dissimilar, the effects of the corresponding phonological match were not statistically reliable. The results suggest that lexical access is nonselective across bilinguals’ two languages, and that the degree of consistency between orthographic and phonological codes influences the manner in which cross-language competition is manifest. Findings are discussed in terms of feed-forward and feed-backward activation dynamics across languages.