Three experiments using rote learning and word-association techniques were carried out to analyze differences in the use of content words (nouns, adjectives, verbs, and adverbs) and function words (pronouns, prepositions, and conjunctions). The purpose of the experiments was two-fold: (1) to extend work on learning, to include function words as well as content words; (2) to analyze the psychological mechanisms that underlie differences in the handling of words from these categories. In Experiment I, paired-associate learning (English word→nonsense syllable, and nonsense syllable→English word) was carried out and differences favoring the content words were established. In Experiment II, an attempt was made to analyze these differences on the basis of the number of associations generated by words from the several grammatical categories. The simple model used gave rise to conflicting results. Experiment III was based on the concept of function words as partially bound forms. It was argued that the function words are incomplete as units and that to have them function efficiently in learning, it was necessary to surround them with other elements. In Experiment III, words from the several categories were surrounded with minimal context, by using triplets (nonsense syllable→English word→nonsense syllable) in a rote learning situation. When the words were embedded in such minimal context, function words were easier to learn than content words. The implications of these findings for the consideration of parts of speech are explored. Analysis on the basis of the subject's customary speech unit is presented as an alternative to analysis on the basis of meaning or meaningfulness.