The purpose of this study was to compare duration characteristics of single-syllable whole-word repetitions (with one and two repeated units) and part-word repetitions (with one repeated unit) in the speech of preschool children who stutter (N = 20) recorded near the onset of their stuttering to those of control nonstuttering children (N = 20). Disfluent episodes were identified in audiotape recordings of the subjects' conversational speech. The digitized signals were analyzed by means of the CSpeech computer software (Milenkovic, 1987). Using visual displays of sound spectrograms, the durations of the spoken repetition unit(s), the silent interval(s) between the units, and the total disfluency were measured. The stutters exhibited shorter silent intervals between spoken repetition units. The duration of the spoken repetition units was very similar for the two groups of children. The total duration of the stutterers' disfluencies was significantly shorter because of their shorter silent intervals when compared to disfluencies of equal repetition units produced by the control subjects. Statistical analysis revealed that silent interval duration was capable of differentiating stuttering from normally fluent children with 72-87% accuracy, dependent upon the disfluency type. To the degree that the groups of subjects represented random samples of the two specified populations from which they were drawn, there appears to be an overall tendency for repetitions during the early stage of stuttering to be produced at a faster rate than repetitions produced by nonstuttering children.