Nonstutterers', mild stutterers', and severe stutterers' acoustic laryngeal reaction times (LRTs) were recorded for isolated vowels and nonpropositional VCV responses in different stimulus conditions governing response preparation. In all stimulus-response conditions severe stutterers produced the longest LRTs, followed in turn by mild stutterers and nonstutterers. The three groups significantly differed from one another in most conditions, but the magnitude of difference between mild and severe stutterers was notably greater than the difference between mild stutterers and nonstutterers. LRT changes as a function of stimulus condition showed that, in general, nonstutterers were best able to use a preparation-facilitating stimulus condition to reduce LRT, and severe stutterers least able to do so. LRT changes as a function of response complexity showed that only nonstutterers produced statistically significant within-group differences. Patterns of LRT change as a combined function of group, stimulus condition, and response type suggest a complex relationship between stutterer severity, preparation time, and type of response complexity. Results illustrate aspects of Goldberg's (1985) model of preparation processes, and support hypotheses that stutterer subgroups show differential preparation deficits along with high motor initiation variability.