Latency, temporal dispersion and input-output characteristics of auditory nerve fiber responses to electrical pulse trains in normal and chronically deafened cat ears were classified and tentatively associated with sites where activity is initiated. Spikes occurred in one or more of four discrete time ranges whose endpoints overlapped partially. A responses had latencies <0.44 ms, exhibited asymptotic temporal dispersion of 8-12 micros and possessed an average dynamic range of 1.2 dB for 200 pulses/s (pps) pulse trains. They likely originated from central processes of spiral ganglion cells. B(1) and B(2) responses (0.45-0.9 ms, 25-40 micros, 1.9 dB) likely stemmed from activity at myelinated and unmyelinated peripheral processes, respectively. C100 micros) likely originated from direct stimulation of inner hair cells, and D8 dB) arose from propagating traveling waves possibly caused by electrically induced motion of outer hair cells. C and D responses were recorded only in acoustically responsive ears. Mean latencies of spikes in all time ranges usually decreased with intensity, and activity at two or even three discrete latencies was often observed in the same spike train. Latency shifts from one discrete time range to another often occurred as intensity increased. Some shifts could be attributed to responses to the opposite-polarity phase of the biphasic pulse. In these cases, temporal dispersion and dynamic range were approximately equal for activity at each latency. A second type of latency shift was also often observed, in which responses at each latency exhibited dissimilar temporal dispersion and dynamic range. This behavior was attributed to a centralward shift in the spike initiation site and it occurred for monophasic as well as biphasic signals. Several fibers exhibited dual latency activity with a 40-90 micros time difference between response peaks. This may have stemmed from spike initiation at nodes on either side of the cell body. Increasing the stimulus pulse rate to 800-1000 pps produced small increases in temporal dispersion and proportionate increases in asymptotic discharge rate and dynamic range, but thresholds did not improve and slopes of rate-intensity functions (in spikes/s/dB) did not change. Responses to high-rate stimuli also exhibited discrete latency increases when discharge rates exceeded 300-400 spikes/s. Spike by spike latencies in these cases depended strongly on the discharge history. Implications for high-rate speech processing strategies are discussed.