This article presents empirical evidence that disfavors using highly lexicalist minimalist models, such as the one presented in Chomsky (1995), when analyzing language mixing. The data analyzed consist of English-Spanish mixed noun phrases discussed in Moro (2014) as well as English-Norwegian mixed noun phrases and verbs taken from the Corpus of American Norwegian Speech. Whereas the lexicalist model in Chomsky (1995) only can explain a subset of the mixing patterns attested in both authentic English-Spanish mixed noun phrases and the American Norwegian corpus, we show that an alternative exoskeletal model can account for all of them. Such a model would entail that rather than assuming lexical items with inherent, functional features that determine the derivation, syntactic structures are generated independently from the lexical items that come to realize them.