Pattern recognition techniques have been applied to the study of structure-taste correlations for perillartine derivatives. The structure of each compound was described by hydrophobicity (log P), logarithm of water solubility (log S), and topological descriptors relating to some positions which were assigned by superposing each compound on a "template" structure. The fragment molecular connectivities were calculated as the topological descriptors. The discriminant functions between the sweet and bitter taste classes were computed by the use of the simplex optimization technique, which correctly recognized most of the compounds under investigation. It was found that the hydrophobicity and one or two topological descriptors concerned with a specific part of the molecules contributed significantly to the discrimination. The discriminant function obtained correctly classified seven of nine compounds (which were not involved in the data set for developing the function) into the taste class to which they belonged.