Abstract: With loudness-adjusted car interior noise as sound stimuli, this work focused on applicability of Nonmetric Multidimensional Scaling (NMDS) to sound quality research. It is presented that NMDS is an effective tool to research subjective assessment of sound quality, when test people is classified by correlation between test people. With this key classification process subjective perceptive structure of stronger consistent car interior noise can be opened out. So it's revealed that:preference can be looked as a main subspace of similarity of car interior sound quality; for both class of test people, preference can be one-dimensional descriptor; for one class, preference has positive relation with descriptor ‘low-frequency’, while for another, preference has negative relation with descriptor ‘low-frequency’.