This paper reports the results of a quasi-experimental effects-of-instruction study examining the efficacy of applying a Cognitive Linguistic (CL) approach to L2 learning of the semantics of English modals. In spite of their frequency in typical input, modal verbs present L2 learners with difficulties, partly due to their inherent complexity — modals typically have two divergent senses — a root sense and an epistemic sense. ELT textbooks and most grammar books aimed at L2 teachers present the two meanings as homophones, failing to address any systematic semantic patterning in the modal system as a whole. Additionally, ELT texts tend to present modals from a speech act perspective. In contrast, CL analyses (e.g., Langacker 1991; Nuyts 2001; Sweetser 1990; Talmy 1988) offer both a systematic, motivated representation of the relationship between the root and epistemic meanings and a rather precise representation of the semantics of each modal. To test the pedagogical effectiveness of a CL account of modals, an effects-of-instruction study was conducted with three groups of adult, high-intermediate ESL learners: a Cognitive treatment group, a Speech Acts treatment group, and a Control group. Results of an ANCOVA indicated that the Cognitive treatment group demonstrated significantly more improvement than the Speech Acts treatment group. The experiment thus lends empirical support for the position that CL, in addition to offering a compelling analytical account of language, may also provide the basis for more effective grammar instruction than that found in most current ELT teaching materials