This investigation was designed to examine the effects of a sound production training program on the production of selected sounds from a speaker with apraxia of speech and Broca's aphasia. Treatment combined modelling repetition, integral stimulation, articulatory placement cueing and minimal contrast practice and was applied sequentially to three sounds in the context of a multiple baseline design. Despite positive acquisition and response generalization effects for all trained sounds, maintenance effects, measured during the training phase of the next targeted sound, exhibited marked declines relative to the level of sound production accuracy that had been achieved during training. A detailed analysis of production errors indicated that overgeneralization of the target sound undergoing training was the most consistent factor contributing to the decline. Preliminary evidence is provided supporting the importance of using multiple sound targets to increase variability during training, and ostensibly, stabilization of learning.