Principal component analysis of the data for 11 characters of 89 cultivars sampled at random from a large collection of Asian origin has reconfirmed the previous assertion by the junior author that the Indica-Japonica diferentiation of rice cultivars is recogni-zable on the basis of character-association patterns, although there is no key character discriminating between the two types. The probability of misclassification by a single character ranged from about 3 percent by the potassium chlorate resistance of seedling to about 40 percent by the length/width ratio of spikelets. Discriminant functions combining 3 or 4 character values gave lower probabilities of misclassification, although there still remained a few intermediate forms having recombined characteristics. The F₁ sterility relationships were almost useless for the purpose of classification. About a half of the total character variance among the cultivars was attributable to Indica-Japonica differenti-ation. The normalized genetic identity between the two types was estimated to be 0.62 from data for 28 alleles at 12 Ioci. A great number of genic differences seem to be involved in the Indica-Japonica differentiation.