In the literature, much attention has been drawn to Japanese noun categorization and numeral classifiers. (Downing 1984, Lakoff 1987, Matsumoto 1991, etc.) In this paper, I argue that Japanese numeral classifiers are syntactic manifestations of number categories which are realized as singular/plural morphemes in non-classifier languages. Mizuguchi(2004) distinguished three types of numeral classifiers: the atomic classifier that takes a count noun (e.g.gakusei san-nin; three students), the collective classifier, and the measure classifier that takes a mass noun (e.g. mizu ni-hai; two cups of water). But the semantic distinction between count and mass nouns in Japanese cannot be based on the homogeneous reference because Japanese nouns are all homogeneous (cumulative and divisive), while English singular count nouns are not. Mass nouns are classified into count mass nouns, such as English furniture and Japanese gakusei on the one hand, and uncountable mass nouns like English water and Japanese mizu on the other hand. Although count mass nouns refer to homogeneous objects such as numeral classifier, like nin, hiki, or dai. Since Japanese lacks singular count nouns and has only mass-type nouns, Japanese nouns cannot be combined directly with a numeral. The numeral classifier attaches to a numeral, building a classifier phrase CLP, and the CLP is inserted into a nominal projection to license the number category of #P (=number phrase) which specifies the number feature of the noun denotation.