Two experiments were carried out to compare disturbance of reading by noises from various sources. Fourteen real noises were selected in one experiment, which were two isolated bursts of sound energy, two quasi-steady impulsive noises, three fluctuating noises, three intermittent noises, two noises from mixed noise sources, one steady noise and the other noise. These noises except for the mixed noises are typical ones classified in ISO 2204 according to level fluctuation. In another experiment, fourteen simulated pink noises were used and their sound levels fluctuated in the same way as those of real noises. L_<eq>, of these noises were equal to about 70 dB(A) and their duration were six minutes. It was found that high disturbance was caused by noises which greatly fluctuated, that is, more contained high sound level even in equal L_<eq> These noises are an aircraft noise, a complex noise which contained aircraft noise and their simulated pink noises. There was high correlation between disturbance and percentile levels like L_1 or L_5 of experimental noises both in real noise and particularly in pink noise. These support the dB(A) peak concept proposed by Rylander et al. It is therefore suggested that unified noise index should contain this finding. In addition, a tentative noise index which could evaluate noises from various sources was shown. The concept of Flindell's pressure L_<eq> was extended and applied to this index. Using this index, it is possible that findings of other researches are reasonably accounted for.