Affective polarization (AP)—the tendency of political partisans to view their opponents as a stigmatized “out group”—is now a major field of research. Relevant evidence in the United States derives primarily from a single source, the American National Election Studies (ANES) feeling thermometer time series. We investigate whether the design of the ANES produces overestimates of AP. We consider four mechanisms: overrepresentation of strong partisans, selection bias conditional on strong identification, priming effects of partisan content, and survey mode variation. Our analysis uses the first-ever collaboration between ANES and the General Social Survey and a novel experiment that manipulates the amount of political content in surveys. Our tests show that variation in survey mode has caused an artificial increase in the mixed-mode ANES time series, but the general increase in out-party animus is nonetheless real and not merely an artifact of selection bias or priming effects.