Complete closure of the glottis is typically treated as the canonical realization of glottal stop, but it has instead been found to be “quite unusual” in running speech. However, such evidence comes mostly from English, with non-phonemic glottal stops. How do glottal stops vary in a language where they are common and contrastive, as in Arapaho (ISO 639 arp)? Does distinctive and frequent use of a glottal stop lead to more canonical productions? Moreover, glottalization is often used to mark prosodic boundaries; Are Arapaho phonemic glottal stops affected by boundary position? Glottal stops in an Arapaho corpus were classified on a scale of lenition and examined for duration, relative intensity, harmonics-to-noise ratio (HNR), and F0. Results show that glottal stops were seldom realized as a stop (only 25%) but instead mostly as glottalization. HNR was lower in glottalization than in adjacent vowels. Word-final glottal stops were more often realized with full closure than word-internal ones. The rarity of full glottal stops in English is also reflected in Arapaho: Greater use of these stops does not result primarily in canonical stop realizations. Moreover, glottal stop realization varies prosodically; thus, glottalization as a prosodic feature is not restricted to non-phonemic glottal stops.