Measuring thresholds for the detection of brief decrements in the level of a sinusoid is an established method of estimating auditory temporal resolution. Generally, a background noise is added to the stimulus to avoid the detection of the "spectral splatter" introduced by the decrement. Results are often described in terms of a temporal-window model, comprising a band-pass filter, a compressive nonlinearity, a sliding temporal integrator, and a decision device. In this study, thresholds for increments, as well as decrements, in the level of a 55 dB SPL, 4-kHz sinusoidal pedestal were measured as function of increment and decrement duration in the presence of a broadband background noise ranging in spectrum level from –20 to +20 dB SPL. Thresholds were also measured using a 55-dB, 8-kHz pedestal in the absence of background noise. Thresholds for decrements, in terms of the dB change in level (L), were found to be more dependent on duration than those for increments. Also, performance was found to be dependent on background-noise level over most levels tested. Neither finding is consistent with the predictions of the temporal-window model or other similar models of temporal resolution. The difference between increment and decrement detection was more successfully simulated by using a decision criterion based on the maximum slope of the temporal-window output.