Despite the similarity in the surface structure, sentences containing a semi-auxiliary (e.g., "avoid,""bother,""happen,""seem,""begin,""tend," etc.) followed by a "to" infinitive or a gerund show a number of differences among themselves in respect to the co-occurrence restrictions imposed on their constituents, certain aspects of semantic interpretation, and the possible range of related sentences. This study examines the characteristics of semi-auxiliaries from these viewpoints and accounts for them in terms of the generative-transformational theory of grammar. A corpus consisting of about one million words of running text of edited present-day American English is examined and the results of the examination are checked with native speakers' comments. The deep structures of semi-auxiliary constructions are determined and semi-auxiliaries are subclassified into six groups on the basis of (1) the types of their underlying subjects, (2) the positions which their embedded sentences occupy in the deep structure, and (3) the restrictions imposed upon the internal structures of their embedded sentences. The following assumptions are made: English has a number of sentence-like elements assigned to different layers of phrase structure by some of the earliest rewrite rules, and not only the initial symbol S, but other types of sentence-like elements can be recursively introduced. (Author/AMM)