The present paper proposes and develops a way of conceptualizing and measuring preference polarization. A society is said to be polarized on preferences if there are significantly sized clusters of individuals with preference orders on alternatives that are reverse of each other. The main theoretical contribution of the paper is an easily implementable technique of extracting the weights of all such clusters embedded in a given preference profile. These weights are used to measure preference polarization. The technique is applied on ballot data from the city council elections for the City of Cambridge, Massachusetts. The analysis of the ballot data provide evidence of increasing preference polarization among the voters.