The evaluation of the perceived spatial quality of an audio system has become more important as the technical possibilities to render spatial information increase. In recent years the field of spatial quality evaluation has been the subject of more thorough investigations than previously, one of the problems being the development of attribute scales appropriate for this purpose. The generation of attributes of spatial audio by means of elements from the repertory grid technique is investigated. In an experiment personal constructs in the form of verbal descriptors were elicited. The constructs were classified by verbal protocol analysis and reduced to a limited number of attributes by cluster analysis. The results show that the repertory grid technique enables a number of attributes of spatial sound quality to be extracted from a group of subjects and that these attributes correspond well with attributes found in other studies. The results also indicate the importance of a definition of the part-the whole or a subset-of the auditory scene to which a specific attribute is referring.