The auditory sensations produced by some AM-tones were studied by evaluating the masking values and based on the experimental results, a masking analog simulator was proposed for the logical estimation of the loudness of AM-tone. A 3kHz-5msec-signal was applied at various phases against a white noise masker whose amplitude was changed with one of the patterns of step, sinusoidal and M-sequence functions. From the viewpoint of the control theory, five non-linearities of AM-masking characteristics were assumed to exist and we reached the following conclusions: The system responsible for the masking of AM-tone can be simulated by a 2nd-order system with the damping factor as the function of its input value. The masking by a monosyllable /ta/ increased markedly at the stage of consonant, which is also proved by our simulator. This analog simulator is valuable for the analyses of the vocal-feature-extracting-process.