The geographic distribution and potential linguistic triggers of the Northern Cities Shift (NCS), a complex chain shift of vowel realizations in urban areas between Madison, WI and upstate New York, have been well-documented (Labov, Yaeger and Steiner 1972; Labov, Ash and Boberg 2006). However, we are left with an actuation problem (Weinreich, Labov and Herzog 1968), especially with respect to social motivations. How might local identity practices (Eckert 1999) relate to similar linguistic processes across such a large area (Labov 2002)? Why should a vowel subsystem that has remained stable for a millenium suddenly shift? Why now, and why should only part of the area with the appropriate pre-existing linguistic system be involved? Why should the shift be absent or delayed among African Americans, rural speakers, or Canadians? This paper proposes a social-historical explanation for the shift: that it was triggered by the Great Migration, the movement from the South into NCS cities of millions of African Americans in the period between 1916 and approximately 1960 (Marks 1989). This population movement, the largest in American history, dramatically changed the ethnic composition of NCS cities. I argue that the first stages of the NCS represented an attempt by white residents to differentiate their speech from that of their new fellow citizens, in effect, a linguistic version of "white flight", the rapid residential segregation that took place in these same cities. Working from 100 years of US Census data and historical descriptions of the Great Migration (e.g., Work 1937), I demonstrate powerful correlations between participation in the NCS and the speed and degree to which communities increased their African American populations, as well as the degree of residential white flight, as indicated by racial segregation and differentiation scales (Mumford Centre 2001). These correlations, paired with the original sound systems of the areas involved, account remarkably well for the temporal, social and geographic boundaries of the NCS, including such distinct features as the exclusion or partial exclusion of Canadians, African Americans, Erie, PA, and rural areas; the eastern and western boundaries of the shift; and the participation of outliers in some other areas, including the St. Louis corridor. I suggest that more detailed city-by-city historical correlative studies might illuminate some of the apparent internal distinctions in the core NCS area, in particular the differences between the highly-focused sound change in western New York state and the apparently more diffuse participation of cities from Cleveland westward. I essay a preliminary theoretical situation of the NCS as an extension of traditionally-invoked social differentiation processes, and suggest ways in which studying the actual processes involved in linguistic white flight might both inform and be informed by work on identity, other changes in progress in American English, the divergence hypothesis (Labov and Harris 1986), and regional differences in African American English (Wolfram 2005).