Abstract. In recent years, a large number of secure voting protocols have been proposed in the literature. Often these protocols contain flaws, but because they are complex protocols, rigorous formal analysis has proven hard to come by. Rivest's ThreeBallot voting system is important because it aims to provide security (voter anonymity and voter verifiability) without requiring cryptography. In this paper, we construct a CSP model of ThreeBallot, and use it to produce the first automated formal analysis of its anonymity property. Along the way, we discover that one of the crucial assumptions under which ThreeBallot (and many other voting systems) operates-the Short Ballot Assumption-is highly ambiguous in the literature. We give various plausible precise interpretations, and discover that in each case, the interpretation either is unrealistically strong, or else fails to ensure anonymity. Therefore, we give a version of the Short Ballot Assumption for ThreeBallot that is realistic but still provides a guarantee of anonymity.