A study investigated children's use of three acoustic parameters (intensity, fundamental frequency, and duration) in the production of two-syllable nonsense words. Subjects were six children each at ages 2, 3, and 4 years with age-appropriate language skills and normal hearing sensitivity. An examiner produced eight novel two-syllable words of similar phonological structure in relation to objects unfamiliar to the children, and then elicited them by imitation. The words were stressed on one or the other syllable. Acoustic analysis of the words elicited suggests that all three of the acoustic features measured are rather poorly controlled at age 2. At 3, both intensity and fundamental frequency appear to be used consistently to mark the stressed syllable regardless of its position in the word. However, it was not until age 4 that the children seemed to have control over duration as a feature for marking word stress. It is also concluded that the data argue for a neutral start hypothesis for early stress production. (MSE)