It is difficult for the multi-source localization method based on time difference of arrival （TDOA） to effectively associate the actual sources between the TDOA values obtained by array microphones. In order to solve this problem， this study developed a multi-source localization method based on TDOA. By using the cross-correlation algorithm to estimate the TDOA values of sound sources， and the locations of multiple target sound sources are solved based on Chan algorithm. In order to eliminate the phantom sound sources， the array microphone is divided into two sub-arrays： localization array and check array. Then the localization-check model for grouped microphone array is constructed. The microphones of localization array are used for locating all possible sources. The microphones of check array are used to eliminate the phantom sound sources and obtain the initial locations of real sound sources. According to the initial locations of real sound sources， a full array TDOA sequences check model is constructed to obtain the final locations of real sound sources. The simulation and experiment platforms are built to verify the proposed method. The simulation and experimental results show that the proposed method can effectively eliminate the phantom sound source based on TDOA， and make full use of array microphones to improve the localization accuracy of multiple sound sources.