Online processing, and the models arising from it, starts with an optimistic view of the American voter, in which it is supposed that the seeming ignorance of voters does not prevent them from expressing rational attitudes about the very political objects they do not know much about. This means that the seeming ignorance of voters is not necessarily a threat to electoral democracy, but the cognitive structures needed for this sort of rationality also lead to necessary, and sometimes extreme, biases in political information processing. Since information stored in long-term memory is linked, both semantically and affectively (that is, based on the perceived positive or negative valence of the information), affect—understood here as a simple positive or negative valence—colors all steps of information processing. For instance, individuals are likely to avoid, or counter-argue, or simply reject information that is at odds with their existing views. As a result, individuals of different political persuasions may have difficulty coming to agreement on the correct interpretation of relevant facts, or even the facts themselves. Alternative memory-based models, which propose that evaluations are constructed on the spot when a question is asked, may help to explain response instability, but fail to serve as complete replacements for the online processing approach. The bias caused by affect-infused cognition seems to present challenges for electoral democracy just as much as the seeming ignorance it accounts for, but it is argued that such biases are mostly limited to individuals who already hold fairly strong existing attitudes, a group which is unlikely to include most voters. Moreover, some degree of intransigence is likely a good thing, as the alternative is views that shift rapidly with new information.