The most fundamental method for judging the sound quality of a product is a subjective listening test. Because subjective tests are influenced by a variety of nuisance variables, systematic statistical data processing is required in order to assure the collection of reliable, reproducible, and statistically meaningful data. In this study, a statistical process consisting of four sequential steps is proposed for handling listening test data obtained by the magnitude estimation method. The first step is to ascertain the reproducibility of the evaluation of each subject for a given sample sound. The second step is to normalize the raw data by adjusting the scale range of individual ratings to fit into a predetermined global scale range. The third step involves determining if any differences exist among the tested sound signals. If differences among sounds are confirmed in the third step, the fourth step is to classify the sounds or sources into groups and to rank the groups. For this statistical process, the analysis of variance and Tukey's method were employed and loudspeakers, having the same dimension and structure but constructed with different diaphragm materials, were used as a model for evaluating sound quality. (C) 2003 Institute of Noise Control Engineering.