Almost of all the sounds around us in everyday life -speech, musical sounds, surrounding noises and others- suffer ever changes in frequency and in amplitude. There are some evidences showing that the changes in frequency convey more important informations than the changes in amplitude do. From this point of view, the thresholds of audibility of frequency change in brief tones with some typical patterns of the changes were measured. With these results obtained and the related facts after other authors, a functional model of detecting mechanism of frequency change was presented. The results obtained are as follows. 1) For three patterns of frequency change, rising ramp, convex triangle and rising step as shown in Fig. 1, the thresholds were determined through the constant method. The initial frequency was 1 kHz. The results are shown in Fig. 2 (a) - (c) where the threshold is defined as the frequency deviation (⊿f) of the variable tone for 75% correct detection. The average results of the subjects show that it becomes harder to detect the change as the duration decreases and also harder to detect the change according to the order of the pattern of rising step, rising ramp and convex triangle in the range of duration (20-300 msec) used. 2) The thresholds were also determined for the pattern of rising ramp and convex triangle for each initial frequency of 250, 1 k and 4 kHz. Two subject's groups participated in the experiment for each pattern of the change separately. The parameters of stimuli are shown in Table 1 and the results are in Fig. 3. Although there are differences in the normalized thresholds (⊿f/f_i) between the patterns of change, the curves of the thresholds, which vary with the duration, for both patterns are similar in each initial frequency. For each pattern with the initial frequency of 4 kHz, thresholds are kept constant for tones of longer duration than about 100 msec. This is not the case with other initial frequencies. 3) The functional model of detecting mechanism of frequency change is shown in Fig. 4. The weighting network in the model is composed of a low pass filter of a simple RC network with a time constant of 13 msec, the inverse transfer function of which is seen in Fig. 5. The comparisons were made of the calculated thresholds by the model with the measured ones in the experiment of 1). The calculated thresholds are roughly the same as the measured ones as in Fig. 6. The comparisons were also made of the results after the model with the results of the experiment in which the frequency change of rising ramp occurred in a portion of the tone burst of 0. 5 sec. The parameters of stimuli used in the experiment are presented in Table 2. Fig. 7 shows that the calculated thresholds and the measured ones have the same inclination although the former ones are lower than the latter ones when the frequency changes occur abruptly near the onset of the burst.