The dramatic differences observed when comparing auditory neural responses to electrical and acoustic stimulation may illustrate one of the important mechanisms underlying the sometimes poor speech recognition abilities of individuals with cochlear implants. Recent research has suggested that the absence of a stochastic component in neural responses to electrical activation may be an important potential mechanism for this degradation in speech recognition performance. There are few psychophysical data, however, demonstrating that this stochastic behavior can be measured directly in implant subjects. In this study, variability in psychophysical threshold was investigated as a measure of the stochastic nature of the underlying neural response in human and non-human subjects implanted with intracochlear electrode arrays. Threshold data collected in both monopolar and bipolar stimulation modes at several phase durations from cat and human subjects are presented. The nature of the neural input/output curve suggests that threshold variability should increase as the slope of the input/output curve is decreased, i.e. as phase duration is increased. These predictions are confirmed by the pattern of psychophysical results measured experimentally in cat and human subjects. Furthermore, the data may suggest that subjects with higher threshold variability, i.e. a relatively greater stochastic component, are more likely to have higher speech recognition scores.