Lists of 24 CVCa nonsense words read by 10 speakers with 10 replications were used to determine speaker specificities in the realization of segment durations. Results indicate that long speech segments (long vowels) display more speaker variability than short segments and that sonorant consonants display less speaker variability than other consonants. Also individual speaking rate proved to be an important factor in this study. Keywords: segmental durations; automatic speaker recognition