We build on the literature on categorization to develop and test a model of input-level spillover effects. Our model predicts that when one firm suffers an accident with an input, investors will punish other users of that input by discounting their stocks, and the magnitude of this negative spillover can be predicted by the non-responsible firm's level of input usage. We also hypothesize that the magnitude of the punishment will be moderated by intermediaries' assessments of the input: ex-ante regulatory sanctions on the input will amplify negative spillovers, while the presence of input-level associations will weaken these effects. Finally, we predict that similarity between the responsible and non-responsible firm on two peripheral attributes—input-portfolio and geographic location—will amplify the negative effect of input usage. We find strong support for our predictions in an event study that examines the stock market valuations of 270 non-responsible manufacturing firms triggered by 78 industrial accidents with a toxic chemical. We highlight our study's theoretical and empirical contributions to the categorization and spillover literatures.