Gap detection thresholds were measured by forced‐choice procedure for conditions where the duration of a silent gap was varied adaptively between pairs of sinusoids of the same or different frequency. Frequencies of the first sinusoid in a pair wereF1= 500, 1000, 2000, and 4000 Hz. Second‐sinusoid frequencies,F2, includedF1=F2, and frequencies 2%, 5%, 24%, and 50% higher thanF1. To minimize the influence of extraneous cues, the onset phase of the second sinusoid in each pair of tones and overall duration of each sinusoid in a pair of tones were randomized. Sinusoids were presented in continuous broadband noise (N0= 30 dB/Hz) at an averageE/N0= 40 dB. Gap detection thresholds increased from 5 to 90 ms as the frequency separation betweenF1andF2was increased. The gap detection thresholds were treated as filter attenuation values by normalizing and converting the data into decibels, and then fitting the roex filter model. On average, the listeners' performances were modeled well by a constant‐percentage (7%) bandwidth filter centered onF1. [Research supported by NIH.]