In three experiments native speakers of Czech translated bare nouns and gender-marked adjective + noun phrases into German, their second language (L2). In Experiments 1-3 we explored the so-called gender interference effect from first language (L1) as observed in previous picture naming studies (naming latencies were longer when the L1 noun and its L2 translation had different genders than when their genders were congruent). In Experiments 2 and 3 we investigated the influence of gender transparency in L2 (longer latencies when an L2 noun has a gender-atypical or gender-ambiguous termination than when its termination is gender-typical). Although both effects were observed in L2 picture naming, only the gender transparency effect could be demonstrated in L1 to L2 translation tasks. The resulting constraints on L2 gender processing during translation are discussed in the framework of bilingual speech production models. (Contains 8 tables.)