This paper compares and contrasts the numeral classifier systems of twogenetically unrelated languages: Newar of Kathmandu and "Hyoujungo",Japanese spoken in Tokyo, the defacto official language of Japan.It is not the case either in Newar or in Japanese that nouns in generalcan be directly quantified by a number. For example, in Newar one cannot say*ni che two house 'Two houses" for 'two houses'. Rather, one must say ni-khache 'two-CL house" Two houses".The noun che 'house' requires that the numeral classifier -kha be affixedto numerals used for counting houses. One might say that che 'house' is notconceived of as a separate countable entity. Only when the unit -kha 'CL' issuffixed to the number do we have a countable entity.Numeral classifiers: The choice of noun determines the choice ofclassifier both in Newari and in Japanese. By looking at the nouns associated witha given classifier it is possible to identify the way in which a noun selects itsclassifier. In Newar there are three different ways in which nouns select theirclassifiers (1) according to semantic parameters (as when sentient nouns select theclassifier -mha, (2) by unique lexical collocation (as when che 'house' selects -kha 'CL') and (3) by full or partial reduplication (as when ha: 'leaf' selects thehead noun ha: itself, where as salakhwa: 'horse's hoof' selects -khwa:, only thelatter portion of the head noun (Hale 2005).These modes can be taken as the basis for a three-term typology ofclassifier selection: (1) semantic selection, (2) lexical selection, and (3)morphological selection. This paper provides a detailed comparison of Newar andJapanese classifiers within each of these three types of selection. Though all threetypes are attested in each language, there are mutual counterparts and skewingsbetween the two languages.