In Experiment I, band noises were used as signals and pure tones were used as maskers. The conventional psychophysical tuning curves (PTCs) present the following characteristics depending on the signal level; slopes of the PTC are steeper than those of the auditory nerve fibers frequency threshold curve (FTC). This result from the large effect of the combination tones and, especially, of the off-frequency listening. Bandwidths of PTC do not depend on both signal levels and duration, but they are narrower than the equivalent rectangular bandwidth (ERB) of auditory filters. Masked thresholds at tips of PTC are affected by the temporal integration of signal sound at signal levels of 10- and 20-dB SL, but are not affected at 30-dB SL. This implies a close relation between PTCs and critical bandwidths (CB-widths). In Experiment II, pure tones were used as signals and band noise were used as maskers. The low- and high-frequency slopes were steeper in Experiment I. The flat parts of the masked threshold at the tips of curves become broader as the signal duration increased. No such flattening was found at a 30-dB signal level, irrespective of signal duration. The results of Experiment II coincide with the prediction derived from the excitation-pattern model. Masked thresholds at the tip of PTC were the same as in Experiment I. The conventional PTC represents the psychological frequency selectivity, and its 3-dB bandwidth corresponds to the CB-width. However, the quantitative properties depend greatly upon the bandwidths of the band noises used as signals or maskers.