1 : The best theoretical account of grammaticality in code-switching is still a hotly debated topic. There are contrasting predictions about what happens at "conflict sites" where the grammars of the two languages have conflicting rules. Compare for example red wine in English with Welsh gwin coch (wine red). In Welsh-English bilingual speech the nominal construction is a potential "conflict site": do we expect e.g. gwin red or red gwin, and coch wine or wine coch? Different theories make conflicting claims on what switches are acceptable. According to the Matrix Language Frame model (Myers-Scotton 2002) word order has to be compatible with the language of the matrix verb. For Cantone and MacSwan (2009), "it is the language of the adjective that determines the position of the NP relative to the adjective".We designed a study to evaluate conflict sites within Welsh-English mixed nominal constructions by using a multi-task approach comprising (1) naturalistic corpus data, (2) an elicitation task, and (3) an auditory judgment task. We discuss the relation between the choice of mixed vs. nonmixed nominal constructions and extralinguistic factors. Our results call into question the usefulness of judgment data for code-switching research and add impetus to the suggestions by Backus (this volume) and Kootstra (this volume) that researchers need to adopt a wider range of methods in this area.