In this paper, we propose to reconcile the controversial debate on Muslim "vote banks" in India by shifting the spatial focus from statewide assessments to the level of constituencies. With the example of Gujarat and Uttar Pradesh in the 2014 general elections, and using an innovative booth-level ecological inference model, we show that Muslims might indeed vote en bloc for or against certain parties, but they tend to do so in a much more localised way than previously assumed. While public Muslim support for the Bharatiya Janata Party did not translate into electoral support in most places, there are important exceptions to this trend - and at least in the case of Uttar Pradesh, their support for competing parties followed a fairly complex spatial pattern. We further explore this spatial variation in Muslim vote pattern by looking at the moderating impact of minority concentration, violent communal history, and ethnic coordination and conclude with a call for more disaggregated research.