Converse (1964) proposed the black-an-white model of attitude stability to describe over-time resonses to repeated questions. The model consist of two groups who are maximally heterogeneous on the crystallization dimension of attitudes. The class of "true attitude" holders provides an identical response at each time period with certainty; the over-time responses of the class of "nonattitude" holders are statistically independent. Previous research employing this model with three-wave panel data has considered all respondents who provided even one "no opinion" or equivocal response as nonopinion holders and combined this group with the estimated nonattitudes under the model. This results in very high levels of nonattitudes. In this research, an argument is developed for treating the nonsubstantive responses probabilitstically. When the tabulations analyzed include an equivocal response category, the simple black-and-white model no longer fits. An alternative black-gray-white model is proposed that fits the data well. Estimates of nonattitudes under this model are markedly lower than under the black-and-white model.