Latent class analysis is used to examine Philip Converse's propositions about political opinion and nonopinion in the American public. Data were taken from the 1956, 1958, 1960 and the 1972, 1974, 1976 American voter panel studies conducted by the University of Michigan Institute for Social Research. Results support Converse's black-and-white model of attitude stability which posits the existence of some stable opinion holders and a group of "nonopinion" respondents whose positions vary unpredictably over time. Rates of opinion holding in the population were estimated to range from a low of 19% for the utilities and housing item, which was Converse's focus, to an atypical high of 59% for the busing item in the second panel. Models allowing for true score change estimated that the number of true score changers was small. The nonopinion respondents were apparently responsive enough to item content that their aggregated responses reflected the climate of opinion among those whose attitudes were better crystallized. As Converse had concluded, there was relatively little overlap of "issue publics"-those who held opinions on one issue were not necessarily opinion holders on other questions. There were not consistent differences in education or political involvement between opinion holders and nonopinion respondents. Relationships of education and political involvement to opinion holding suggest that respondents hold stable opinions on those issues that have greatest personal relevance for them.