Using stepwise multiple regression, voting behavior was predicted from semantic differential scale evaluations of the winning candidate in four voting situations: the 1972 Presidential race in Illinois and Missouri and the gubernatorial races in the two states. "Image" was defined in terms of the semantic differential scales, and "image voters" were defined in terms of the accuracy of the vote prediction. The four sets of communication variables were amount and type of campaign information each respondent claimed to receive from media and interpersonal sources, perceived influence of information sources, believability of media and interpersonal sources, and perceived usefulness of the sources for seeking political information. The findings suggest that while certain instances of"television image voting" were isolated, the phenomenon does not seem to be a general rule in assessing voting behavior. Party loyalty and political party preference was the largest factor relating to how people voted in all four voting situations. The prediction of vote from semantic differential scales was weak and it is suggested that the semantic differential scales be reassessed. (Author/TS)