A common language ideology in the United States is that New York City English (NYCE) displays reliable geographic variation across the city’s five boroughs, what we call the Borough Accent Ideology (BAI). In direct contrast, linguists argue that borough accents do not exist, but instead serve as a proxy for socioeconomic differences in NYCE (Hubbell 1950, Bronstein 1962, Labov 1966, Labov, Ash, and Boberg 2006:234). This paper contributes the first empirical evidence related to the BAI, with an analysis of perceptual data from an interactive website where listeners heard short audio samples of native New Yorkers and assigned them to one of the city’s five boroughs. The results confirm that listeners cannot accurately discern a talker’s borough of provenance, but also that listeners are not guessing when they vote. Based on the descriptive patterns, we hypothesized that listeners create a binary opposition between Manhattan, which is the borough that is least-aligned with traditional NYCE, and the outer boroughs, where listeners expect to hear higher rates of NYCE features. A regression analysis confirms this hypothesis, and finds specifically that a talker’s use of variable non-rhoticity and BOUGHT-raising are significant predictors of votes, with more rhoticity and less-raised BOUGHT predictive of votes for Manhattan. In addition, there is no significant difference between native and non-native New Yorkers in voting behavior, suggesting that this binary strategy is accessible to speakers from both within and outside New York City. Overall, the results confirm that the BAI remains an ideology and not a linguistic reality, at least for the task in question.