The incorporation of low levels of noise into an electrical stimulus has been shown to improve auditory thresholds in human subjects. In this paper, thresholds for noise-modulated pulse-train stimuli are predicted by utilizing a stochastic neural-behavioral model of ensemble fiber responses to bi-phasic stimuli. A neural spike count comparison rule has been presented for both threshold and intensity discrimination under the assumption that loudness is a monotonic function of the number of neuron spikes. An alternative approach which we have pursued involves analyzing the neural response to each individual pulse within a pulse train to investigate the threshold behavior. The refractory effect is described using a Markov model for a noise-free pulse-train stimulus. A recursive method using the conditional probability is utilized to track the neural responses to each successive pulse for a noise-modulated pulse-train stimulus. After determining the stochastic properties of the auditory nerve response to each pulse within the pulse train, a logarithmic rule is hypothesized for pulse-train threshold and the predictions are shown to match psychophysical data not only for noise-free stimuli but also for noise-modulated stimuli. Results indicate that threshold decreases as noise variance increases.