This paper proposes three scaling methods for qualitative three-mode three-way data. A real data set collected to investigate the impression of pictures by the semantic differential (SD) method is analyzed as numerical examples. The data consists of three modes ; raters, SD items and pictures. All the models classify I items into G groups to construct uni-dimensional scale in each group by introducing fuzzy c-means criterion into homogeneity analysis. In the first model, each scale score is expressed as a function of a design matrix. The data is analyzed assuming that scale scores of all raters to the same picture are equal. Appropriate scales are, however, not constructed, indicating that there are differences among the raters. In the second model, rank of scale scores is restricted to Rg in order to explore the differences of raters. In the third model, raters are clustered into D groups to find which raters are different. The selection of Rg or D is perfomed by means of increasing the parameter value until appropriate classification is obtained.