Previous evidence has shown that word pairs that are either related in form (e.g., ruc-berro; donkey-watercress) or very closely semantically related (e.g., ruc-caballo, donkey-horse) produce interference effects in a translation recognition task (Ferré et al., 2006; Guasch et al., 2008). However, these effects are not observed when the words have a less close semantic relation (e.g., ruc-oso, donkey-bear). The lack of interference in less similar words could be due to the low level of activation of the corresponding semantic representations by the time the translation decision has to be made. The present experiments tested this possibility using the same materials as the previous studies but decreasing from 500 ms to 250 ms the presentation time of the word to be translated. Performance of highly proficient bilinguals of Spanish and Catalan was examined in two experiments. Catalan-Spanish translation direction was tested in Experiment 1 and Spanish-Catalan direction in Experiment 2. The results showed significant effects only with form and very close semantic relations, but not in the case of less closely semantically related words. The pattern of results was the same, regardless of translation direction and language dominance.