The preferences for the qualities of sounds reproduced through the system varying in frequency characteristics were investigated on the basis of a vector model in multi-dimensional space. The model assumes that the preferences for a sound quality might be determined by the scalar product of two vectors in the space, one for the sounds and the other for the listner and program-sources. The main findings are as follows; (1) The preference space for sound qualities could be described as the real Euclidean space of four orthogonal dimensions. (2) Those dimensions were interpreted as treble, bass, balance, and noise. (3) The listeners (judges) and program (music) sources (A), and frequency characteristics of the reproducing system (Z) were mapped as vectors in the space. (4) There seemed to be some interactive relation between them. The listeners prefered such characteristics that could reproduce some distinctive features of program-sources. They were divided into four types: i.e. high-bass*low-treble, low-bass*high-treble, balance, and medium-tone types. (5) Listners were also divided into three sub-groups. One of them was named "absolute", because listeners in this group showed an absolute preference for one specified type of characteristic; the second one "selective", because those in this group prefered two or three characteristics selectively according to the program-sources in question; and the last "decentralized", because the preferences of those in this group exhibited a tendency of much too discursive selection. (6) The validity of the vector model was tested. It ensured that the model could predict the experimental results fairly accurately.