The present study addresses issues related to the romanization of Modern Standard Arabic (MSA) words pronounced with an initial glottal stop, both through transliteration and phonemic or phonetic transcription. The study argues that, in MSA, an utterance-initial glottal stop which is an integral part of a word or a prefix is phonemic, while an epenthetic glottal stop is non-phonemic, and that each type should be represented differently in romanization. This argument is supported by evidence from near-minimal pair tests and comparisons between MSA words with epenthetic glottal stop and English vowel-commencing words, as well as insights from cognitive phonology. In this context, a number of authentic transliterations and transcriptions, drawn from widely used romanization systems as well as essential works in phonetics and Arabic language teaching, are examined to see how word-initial glottal stops are romanized in different utterance positions. This has revealed a number of recurrent problems in representing word-initial glottal stops as well as the words immediately preceding them when they occur in utterance-medial position. It is suggested that most of these problems could be avoided by observing the distinction between phonemic and epenthetic glottal stops.