This article presents a cross-language study of voicing in initial stops, focusing on the acoustic measurements of these sounds. The authors investigate how the timing of voice onset relative to release can serve as a basis for distinguishing different manner categories of stops in various languages. The study involves spectrographic analysis of high-quality recordings made in an acoustically treated room, with informants producing words and sentences containing initial stops. The data are analyzed for both isolated words and sentences, and the results are compared across eleven languages: Dutch, Puerto Rican Spanish, Hungarian, Tamil, Cantonese, English, Eastern Armenian, Thai, Korean, Hindi, and Marathi. The study finds that voice onset time generally serves as a reliable measure for separating stop categories, though there are some cases where it is insufficient. For example, in Hindi and Marathi, the voiced unaspirated and voiced aspirated stops show systematic differences in average values but occupy overlapping ranges. The velars tend to have higher values than other stops, and the distribution of values is sensitive to the place of stop closure. The overall frequency distributions of voice onset time values are tri-modal, with modes centered at −100, +10, and +75 msec. In sentences, the voice onset time dimension still effectively separates stop categories, though there is some effect of embedding stops in running speech. For categories with voicing lead, voicing often proceeds unbroken from a preceding voiced environment into the closure interval. The study concludes that while voice onset time is a useful measure, it is not perfect and can be influenced by factors such as the presence of other sounds before the stop.This article presents a cross-language study of voicing in initial stops, focusing on the acoustic measurements of these sounds. The authors investigate how the timing of voice onset relative to release can serve as a basis for distinguishing different manner categories of stops in various languages. The study involves spectrographic analysis of high-quality recordings made in an acoustically treated room, with informants producing words and sentences containing initial stops. The data are analyzed for both isolated words and sentences, and the results are compared across eleven languages: Dutch, Puerto Rican Spanish, Hungarian, Tamil, Cantonese, English, Eastern Armenian, Thai, Korean, Hindi, and Marathi. The study finds that voice onset time generally serves as a reliable measure for separating stop categories, though there are some cases where it is insufficient. For example, in Hindi and Marathi, the voiced unaspirated and voiced aspirated stops show systematic differences in average values but occupy overlapping ranges. The velars tend to have higher values than other stops, and the distribution of values is sensitive to the place of stop closure. The overall frequency distributions of voice onset time values are tri-modal, with modes centered at −100, +10, and +75 msec. In sentences, the voice onset time dimension still effectively separates stop categories, though there is some effect of embedding stops in running speech. For categories with voicing lead, voicing often proceeds unbroken from a preceding voiced environment into the closure interval. The study concludes that while voice onset time is a useful measure, it is not perfect and can be influenced by factors such as the presence of other sounds before the stop.