Prior research has investigated the quality of information a reader can extract from the parafoveal word to the right of the currently fixated word. However, very few studies have considered parafoveal processing in bilingual readers, who may differ from monolinguals due to slower lexical access — particularly in the weaker language — and susceptibility to cross-language activation. This eye-tracking experiment therefore investigated how bilingual readers process parafoveal semantic information within and across native and non-native languages. We used the gaze-contingent boundary technique in which a preview word embedded in a sentence was replaced with a different target word when the eyes crossed an invisible boundary prior to the critical word. Before the boundary change, parafoveal preview words appeared either in the same language as the rest of the sentence (non-switch) or in the alternate language (code switch). Upon fixation, target words always appeared in the same language as the rest of the sentence to create an essentially monolingual language context. Semantic relatedness was manipulated between previews and targets (synonym/translation vs. unrelated previews). Semantic preview benefits emerged for non-switched synonym previews but not for code-switched translation previews compared to unrelated previews. Furthermore, participants skipped over code-switched previews less often than non-switched previews and no more often than words that were unfamiliar to them. These data suggest that bilingual readers can extract within-language semantic information from the parafovea in both native and non-native languages, but that cross-language words are not accessible while reading in a monolingual language mode, as per the partial selectivity hypothesis.