We report the results of a systematic study of the feasibility of automatically classifying documents by genre using adjectives and adverbs as indicators of affective language. In addition to the class of adjectives and adverbs, we focus on two specific subsets of adjectives and adverbs: (1) trait adjectives, used by psychologists to assess human personality traits, and (2) speaker-oriented adverbs, studied by linguists as markers of narrator attitude. We report the results of our machine learning experiments using Accuracy Gain, a measure more rigorous than the standard measure of Accuracy. We find that it is possible to classify documents automatically by genre using only these subsets of adjectives and adverbs as discriminating features. In many cases results are superior to using the count of (a) nouns, verbs, or punctuation, or (b) adjectives and adverbs in general. In addition, we find that relatively few speaker-oriented adverbs are needed in the discriminant models. We conclude that at least in these two cases, the psychological and linguistic literature leads to identification of features that are quite useful for genre detection and for other applications in which identification of style and other non-topical characteristics of documents is important.