Bilinguals commonly draw on their two languages within a single speech event, a practice known as code-switching. On the basis of different methods and bilingual populations, various theoretical accounts of code-switching have been developed. Yet, while theories proliferate, cross-fertilization between them remains limited. Hence, the question that guides this paper is: how can we better understand the nature of mixed interactions, with a view to creating more accurate models of (multilingual) language competence? We show how a multimethod, comparative approach that integrates linguistic, psycholinguistic and social factors will help us draw a distinction between which codeswitching patterns are uniform across communities and language pairs, and which patterns are variable. Addressing both the nominal and verbal domains, we present findings from a series of comparative studies that investigate how bilinguals from different communities produce, judge, or process bilingual structures. We discuss to what extent bilinguals (i) produce strings that can be seen as having the same syntactic structure within and across communities, (ii) make the same linguistic judgments, and (iii) converge in their processing of these strings. We highlight the importance of surveying the patterns that emerge across communities, rather than on an example and counterexample basis, in order to bring our understanding of code-switching, and of language as a whole, forward.