Measurements of the just-noticeable change in resonance frequency ∼(ΔF)r of a second-order filter are reported. The source signal was either periodic, with or without a smooth change in fundamental frequency, or it was random white noise. These differences in the nature of the source had little effect on ∼(ΔF)r. Over the investigated ranges of reference resonance frequency (Fr=300 to 2000 Hz) and filter selectivity (Q=1 to 36), the results are well summarized by ∼(ΔF)r=0.079 Fr/(Q)1/2. The data were used to evaluate filter-bank models employing different filter shapes and performance-prediction schemes. A good fit to the resonance-discrimination data was obtained with filters derived in a masking study by Patterson [J. Acoust. Soc. Am. 55, 802–809 (1974)] and with performance based on use of the maximum level difference over all bands. The latter finding indicates that listeners may not make optimal use of small level differences distributed over multiple bands.