This article addresses the theoretical and practical challenges in classifying polysemous terms within specialized terminological systems. Drawing on foundational research in cognitive semantics and terminology theory (Apresjan, 1974; Cabré, 1999; Cruse, 2000; Pustejovsky, 1995), it reviews existing classification models, examines their limitations, and analyzes empirical cases from engineering, medicine, and information technology vocabularies. The study proposes an integrated framework that combines sense-relation typologies with corpus-driven metrics to enhance consistency and usability in terminological resources. Implications for lexicography, terminology management, and automated term-disambiguation systems are discussed.