This paper presents part of the results of an experiment on information processing during simultaneous interpretation (SI). The traditionally dominant theoretical approach to SI rested on the idea that interpreters consistently struggle in their search for the deep meaning of the source speech, which is then reformulated in the target language in a natural fashion. The process of SI was supposed to be, therefore, identical to that of ordinary human communication, where utterances are originated from spontaneous thought. Moreover, interpreters do not translate utterances, they interpret them. Experimental results from various independent researchers, however, appear to point to the intervention of a double system, with an alternation between the search of deep sense and the immediate transliteration of units or more complex utterances when the relation between source and target language so allows. The author claims that given the pressing demands imposed by SI, interpreters behave strategically and use the most effi cient type of processing for each unit in order not to exhaust their already limited cognitive resources. This may have serious implications for the training of interpreters.