The up-down method of adaptive psychophysical measurement uses binary response categories, e.g., “stronger” and “weaker.” This study proposes that ratings using three response categories, e.g., “stronger,” “do not know,” and “weaker,” or four response categories, e.g., “stronger,” “probably stronger,” “probably weaker,” and “weaker,” should be used instead. Simulation experiments showed that the proposed methods were superior to the standard up-down method. Comparisons were made with respect to the root mean square error (RMSE). First, in the case of two response categories, the RMSEs of estimates made using a stochastic model were smaller than those derived using the standard arithmetic method based on simple averaging, except in one extreme case. Hence, comparison of two, three, and four response categories was made with respect to estimates made using stochastic models. The RMSEs of estimates of the point of subjective equality using three or four response categories were smaller than those using two response categories. The RMSEs of estimates of model slope parameters, where a just noticeable difference was calculated as a ratio of the parameter, were smaller with three or four response categories than with two response categories, except in two extreme cases.