Negative voting occurs when voters respond more strongly to political actions or outcomes they oppose than to comparable actions or outcomes they favor. This paper discusses the possibility that negative voting is an artifact. We develop a simple probabilistic model of constituent support that produces negative voting among some constituents purely as a function of the boundedness of probabilities. The model has two implications that will help to determine whether empirical findings about negative voting are real or artifactual. First, negative voting should characterize the behavior of a politician’s previous supporters, whereas previous opponents should show positive voting. Second, negative voting should be stronger the higher the previous level of support for a politician. Given that incumbents almost by definition have more previous supporters than opponents, empirical studies should be more likely to find negative voting when analyzing the vote for incumbents, and negative voting should be more noticable the higher the incumbent’s previous level of support. An analysis of voting in 1982 and 1986 House races produces evidence consistent with these propositions.