The ultrasonic pulse echo method was adopted to test the localized void defects in thick section composites and the extracted ultrasonic backscattered signal was analyzed. The results showed that the resonance structure noise may appear in the near surface region of the material. The voids can result in frequency reduction and amplitude decrease of the structure noise. The backscattered signal was back to normal state in the far surface region and the voids can lead to the appearance of low amplitude echo. The localized void identification method based on the backscattered signal processing was proposed. The near surface signal was processed by a modified S-transform to generate time-frequency coefficient matrix. The minimum and gain of the signal main frequency can be calculated to identify the voids. For the far surface signal, the wavelet transform modulus maximum denoising method was applied firstly, and then the denoised signal was processed by the modified S-transform to generate time-frequency coefficient matrix. Finally, the maximum and gain of the signal main frequency could be calculated to identify the voids. The experiments were implemented for the thick section composite plate specimen and curved surface specimen, respectively. The results showed that the localized void defects in thick section composites could be effectively identified by this signal process method.