What role do parties play in electoral competition? Why would parties exert discipline on their members? This investigation links these issues theoretically and empirically by modeling parties as informative brands to voters. As conventional wisdom suggests, most party labels carry fairly precise meanings. Democrats are generally liberal, Republicans are generally conservative, and Libertarians are basically conserva- tive but often socially liberal. The theoretical question is, how is this sustained as an equilibrium phenomenon? In the basic model, the researchers assume there are a large number of constituencies, voters are risk averse and incompletely informed about candidate ideal policies, and candidates are unable to commit to a declared policy platform. In this environment, parties can play a critical role by ag- gregating ideologically similar candidates. Importantly, party membership is endogenous. Certain types of candidates thereby have an incentive to affiliate themselves with a party to reduce voter uncertainty about their preferences. The results of a pilot by the investigators are the following. (1) Parties can be effective screening devices for candidate preferences. Intuitively, this screening works because party membership imposes differential costs across candidate ideological types, so that only candidates whose preferences are sufficiently close to a party 's platform are willing to affiliate. Even in a single party system, voter welfare is increased by the information party candidates are able to convey relative to unaffiliated candi- dates. (2) In a two party system there is a direct link between the effectiveness of screening and platform choices. When the costs of joining parties is high-e.g., parties can effectively discipline their members or screen out certain type of candidates -party labels are very informative, and the parties' platforms converge in Downsian fashion. When party labels are less informative, however, the platforms diverge.This happens because taking an extreme position reduces a party's ideolog- ical heterogeneity, thus making its label more meaningful to voters. (3) Candidates do not always affiliate with the nearest party, since they also take the electoral benefits of party membership into account. These benefits may vary depending on constituency preferences. Thus even candidates with the same ideal policy may affiliate with different parties, depending on the preferences of their district's median voter. Extensions of the basic model provide additional predictions. Of particular interest is the possibility of hot races in which voters learn much more about candidates than we assume in our basic model. This extension can account for the fact that party labels sometimes mean different things in different places. Another extension is to model parties as democratic institutions that must choose platforms and screening mechanisms collectively, perhaps through a primary system. Others include endogenizing the number of parties, exploring alternate electoral systems, and determining the optimal level of screening or discipline. Both the initial results and the proposed extensions are amenable to empirical analysis. Among the topics we will pursue are the effectiveness of party branding, the relationship between party screening and platform choices, retirement and defection patterns, and the evolution of platforms over time.