This dissertation investigates compound tensification in Korean by examining the phonetic realization of word-medial stops through production and perception. Prior research on Korean stop phonetics has focused predominantly on word-initial position, leaving word-medial stops comparatively understudied. This is a gap that warrants attention given that characterizing word-medial stop realization is prerequisite to accounting for the actual realization of compound tensification, a process whereby a plain stop surfaces as tense at the onset of the second noun in a noun-noun compound. Most previous studies have investigated this phenomenon through dictionary entries and speaker surveys rather than direct phonetic measurement, and have treated tensification as categorically present or absent. This approach obscures how derived tense stops behave phonetically relative to underlying plain and tense stops, questions that bear directly on the phonetics-phonology interface. The present study provides a systematic acoustic study of 32 speakers, as well as a perceptual study of 94 listeners. The acoustic study examines phonetic realization of word-medial stops across six phonetic dimensions, including closure duration, voicing proportion during closure, positive VOT, burst amplitude, F0, and H1*-H2*, to evaluate four sets of hypotheses. First, the Lenition and Neutralization Hypotheses predicted positional weakening and full merger, respectively, of stop contrasts in word-medial position. Second, the Categorical and Gradient Realization Hypotheses predicted full versus incomplete acoustic convergence of derived tense stops on underlying tense stops. Third, the Invariable and Variable Realization Hypotheses predicted uniform versus speaker- and item-specific phonetic outputs and are evaluated through speaker- and item-level analyses. Finally, the Morphological Boundary Enhancement Hypothesis predicted enhanced acoustic contrasts at compound boundaries. The perception study then examined how native listeners map these acoustic correlates to stop category judgments. The key findings are as follows. The phonetic patterns of derived tense stops align with the Categorical Realization Hypothesis whereby their realizations either overlap with underlying plain stops in simple nouns or underlying tense stops in simple nouns. Also, compound tensification is variably realized. Speakers show systematic individual differences alongside considerable within-speaker variation, while item-level variation remains only partially explained by word familiarity rating. No evidence is found for lenition or neutralization in word-medial position; instead, a redistribution of acoustic correlates is observed whereby stop contrasts are realized primarily through closure duration. Morphological boundary enhancement is restricted to F0 in select contexts, providing only weak support for a general enhancement effect. In the perception study, listeners relied on closure duration to distinguish plain and tense stops. Listeners tracked the acoustic consequences of tensification without access to morphological information, while confidence ratings suggest sensitivity to the listeners’ awareness of the phonological context. These results are discussed in terms of the relationship between phonetic variation and phonological contrast, particularly how differences in gestural organization are reflected differently across acoustic correlates, and the relationship between speech production and perception.