Rhythm of speech in the literature of the past decade is quantified using measures of durations: first order, such as mean duration and percentage of the total sample, or second order, such as variance and pairwise variability index. Results of different studies, however, vary even when the same materials and speech mode are used and imply uncertainty of cross-study comparisons. Allen (1978) discussed the issue of validity (the criteria used for segmenting) and reliability (how accurately these criteria can be applied) on duration studies. The current paper examines how different segmentation criteria affect the five most commonly used rhythm metrics. All metrics are computed on the segmented sample of the read story (The North Wind and the Sun) for three speakers each of American English, Indonesian, and Serbo-Croatian. Different segmentation criteria were applied to (1) borders between voiceless plosives and vowels and (2) borders between vowels and trills. As a result, durations of some segments were changed in the range of 10–50 ms. The results show how different metrics are affected by the choice of two segmentation rules at two different speech rates. Implications for cross-linguistic studies are discussed.