In this paper we ask whether individuals decide that people are generally trustworthy or untrustworthy by extrapolating from their experiences in localized interactions or whether a more fixed predisposition drives assessments of trustworthiness. These two contrasting theoretical perspectives on generalized trust can be translated into empirically testable models and adjudicated using confirmatory tetrad analysis. This paper is among the first substantive applications of confirmatory tetrad analysis and illustrates an important advantage of this technique--the ability to distinguish between causal and reflective indicators of a latent variable. We find that individuals develop a generalized expectation of trustworthiness based on their experiences with different groups of people in localized settings. We demonstrate the robustness of our results across disparate samples and spatially dissimilar survey sites. (Contains 8 footnotes, 2 tables, and 1 figure.)