The present study investigated a theory of the etiology and generalization of stuttering behavior. The subjects were 24 male students at a medical center who responded to advertisements requesting participation in a research project on learning and heart rate. The age range of the subjects was 22-28 years, and the mean age was 22.8 years. Three stimulus topic words were used in the present study. Two of these words were semantically equivalent. The independent variables were the three words that the subjects were instructed to discuss: neutral word condition, experimentally induced noxious word condition, and word semantically equivalent to the noxious word condition. The six dependent measures were percentage of part-word repetitions, percentage of interjections, heart-rate beats per minute, electromyographic microvolts per minute, galvanic skin response ohms conductance per minute, and self-report ratings of "state" anxiety. As predicted, heart rate and a self-report "state" anxiety measure achieved significance in a pattern corresponding to part-word repetitions. Two other measures of state anxiety (galvanic skin response and frontalis electromyographic activity) did not reach significance, although the results were in the predicted direction. The theory that "trait" anxiety accounts for susceptibility of fluency failure under emotional stress was not confirmed. The hypothesis offered here (that part-word repetitions generalize along a semantic dimension) received only tentative support. Discussion of the results centered around the demonstration in the present study of the need for multifaceted treatments tailored to the individual learning histories of specific stuttering behaviors, as well as molecular assessments of stuttering prior to treatment.