In this paper, an attempt was made to identify and optimize the rattle noise signals by relative approach based on the hearing characteristics of human ear. In the first step, the noise signals at the driver’s right ear were filtered by the transfer functions of the outer, middle and inner ear, and the steady noise signals were extracted by the regression and smoothing processes. In the second step, the gear rattle noise relative approaches were obtained by subtracting the steady noise signals from all of the filtered noise signals. The gear rattle noise relative approaches were the transient noise signals of the unloaded gear pairs which caused the gear rattle noise. In the last step, the occurrence time, frequency and level of the gear rattle noise were obtained from the relative approaches. The result of the gear rattle noise identification was consistent with that of the subjective perception and identified accurately. The rattle noise contribution of all the unloaded gear pairs was obtained by using the gear rattle noise relative approaches. The results show that the reverse gear pairs were the main source of the gear rattle noise. The rattle noise level was decreased by 17.4% with the backlash of the reverse gear pairs being reduced by about 50.0% by enhancing the gears process precision, and the rattle performance was effectively improved.