A B S T R A C TFollowing recent work that questions traditional social categories, this paper examines the role of ethnicity in conditioning linguistic variation. Reporting on a large-scale project in the multicultural context of Toronto, we argue for combining emic and etic approaches to social categorization. Focusing on the Chinese and Italian communities, our analysis of two sociolinguistic variables shows that speakers may vary in overall rate, but linguistic conditioning remains largely constant across and within ethnic groups. Whereas there is evidence for language transfer in the first generation, differences between generations suggest that transfer does not persist. Some speakers appear to use overall rates to express ethnic identity. Differences between communities may be explained in terms of different timelines of settlement and visible-minority status.A fundamental question of sociolinguistics is how social differences among speakers are reflected in their linguistic behavior. In the study of sociolinguistic variation and change, this question has been approached predominantly through "social grouping" (Horvath & Sankoff, 1987:179): that is, assign speakers to previously defined categories, and then correlate those categories with quantitative differences in linguistic behavior. A small set of social categories has been correlated with linguistic variables: sex, age, social class, and ethnicity. All of these categories featured in Labov's pioneering work on Martha's Vineyard (1963) and in New York City (1966) and have been replicated in many studies. As a result, we now have a good descriptive understanding of how each of these categories is implicated in linguistic variation: to Walker. We thank Walt Wolfram, whose invitation to participate in a panel discussion prompted Walker to start thinking systematically about ethnicity (Walker, 2001), our fieldworkers and research assistants for their hard work, Bryn Greer-Wootten for his advice on statistical techniques, and most of all the people of Toronto for sharing their time, reminiscences, opinions, and stories with us. Previous versions of different parts of this paper were presented between 2005 and 2007 at various international conferences. We thank the audiences for their comments and several anonymous reviewers for their feedback. We especially thank Barbara Johnstone for helping to clarify the discussion of the results. Any remaining errors are our own responsibility. Authors' names are listed alphabetically.