Western Amazonian languages stand out in having classifiers that - in addition to the well-established classifier environments - also appear as derivational devices on nouns. Since classifiers are commonly assumed to originate in nouns, classifier languages confront us with an analytical problem in the domain of binominals, i.e. how to distinguish between the derivational use of classifiers on nouns, and noun-noun compounds. The present paper addresses this problem on the basis of primary data from Harakmbut (isolate, Peru) and Mojeno Trinitario (Arawak, Bolivia), two unrelated Western Amazonian languages. As a factor bearing on this problem, we show that in both languages the noun/classifier distinction is blurred by the fact that there is a class of nouns that share many features with the canonical classifiers, i.e. that of bound nouns. In this paper, we discuss how noun-classifier derivation differs from noun-noun compounding, or classifiers from bound nouns for that matter, at the phonological, semantic and syntactic levels in both languages.