The detection of an intensity increment in a longer duration sinusoid or pedestal is often used as a measure of intensity resolution, but the decision processes underlying this measure are poorly understood. Thresholds were obtained for detection of an increment in a 370-ms, 4-kHz pedestal in quiet or in noise to determine the relative contributions of background noise level and pedestal level, the effect of increment duration, and the effect of different noise spectra. Increment detection thresholds expressed in units of DeltaL[10 log(1+DeltaI/I)] decreased as pedestal levels increased. At low pedestal levels, increment detection was limited by the masking effect of the noise and was similar across noise conditions for pedestals of equal sensation level. At high pedestal levels, the noise had no effect and increment detection was determined by the pedestal level in dB SPL (sound pressure level). Increment detection improved with increasing increment duration and was altered less by a noise band above the pedestal/increment frequency than by a broadband noise that produced equal masking at the pedestal/increment frequency. The quadratic-compression model described by Neely and Jesteadt [(2005). Acta Acust. Acust. 91, 980-991] provided a better approximation to the data than a model based on excitation patterns.