The capacities of the goldfish to detect changes in sound burst repetition rate were studied using classical respiratory conditioning. In experiment I, the just detectable amount of an instantaneous random jitter of burst period was measured as a function of mean period. In experiment II, the just detectable amount of slow sinusoidal jitter of burst period was measured for three burst types having different spectral and waveform characteristics. In experiment III, sinusoidal jitter detection thresholds were measured in the presence of varying degrees of random jitter. The data show that (1) rms sinusoidal and random jitter in a periodic train of bursts are approximately equally detectable. (2) Sinusoidal jitter detection depends upon burst period duration and the short-term envelope definition and not upon frequency-domain information. (3) For a given burst period, stimulus jitter and an internal temporal noise appear to add independently to determine period discriminability. (4) Psychophysical estimates of internal temporal noise are 0.160 and 0.710 msec at periods of 5 and 10 msec, respectively. The data are consistent with the hypothesis that the burst period discrimination task is based upon a measurement of the time interval (duration) between spikes in auditory neurons.