Previous research has suggested that there is a cost in switching between language in a lexical decision task. This paper reports two studies exploring its basis. Experiment 1 confirmed such a cost in a lexical decision task in which the target language for a trial is specified. German-English bilinguals were slower on switch trials compared to nonswitch trials. Experiment 2 changed the task to one in which a word response could be given independent of language. In this case, there was still a cost of switching between languages but it was much reduced. We propose that in bilingual lexical decision a cost arises because of the process of testing the output of the lexico-semantic system. Different lexical decision tasks induce different output tests. We also outline a mechanism that could give rise to these effects based on the notion of competition between language schemas.