The psychophysics and neurophysiology of sound detection in quiet and under noise masking were studied in goldfish. Psychophysical masking is a linear function of masker level. For long duration signals, signal-to-noise ratios (S/N) at threshold are 15.5, 19, and 22.5 dB for 200, 400 and 800 Hz signals, respectively, and is -5 dB for a noise signal. Threshold declines with signal duration to about 700 ms. The slopes of the masked temporal summation functions are about unity, indicating that energy is constant at threshold. In quiet however, the slopes are generally less than 0.5, indicating that shorter signals are detected at lower energy. Neural correlates of the masked S/Ns and the slopes of temporal summation functions were sought in the response patterns of single saccular neurons. Rate- and synchronization-intensity functions were obtained for tone and noise signals in quiet and in noise. S/Ns at behavioral threshold correspond closely to those required to raise spike rate just above that evoked by the masker alone, but are well above those required to cause clear synchronization. Therefore, sound detection is probably based on spike rate and not synchronization criteria. The equivalence of behavioral and neural thresholds indicates that the filters used in behavioral sound detection are simply the bandwidths of saccular fibers. A model outlined by Zwislocki which predicts the rate of temporal summation from the rate of growth of neural activity with intensity accounts quite well for the observed slopes of temporal summation functions both in quiet and in noise.