The invention provides a coding unit partition method. The method includes the following steps: A1, inputting more than two viewpoint video texture images and depth image sequences, A2, recording partition information of all maximum CUs of independent viewpoints, A3, establishing mapping between the partition information of the maximum CUs and weights, A4, acquiring matching units of the maximum CUs in the independent viewpoints, A5, determining division of sub-blocks corresponding to the maximum CUs to be coded in texture images of current non-independent viewpoints to be coded, and A6, outputting division of all the maximum CUs of current coded viewpoints. According to the CU partition method, statistical information of division of the CUs of the coded synchronous independent viewpoints is used for achieving division of the CUs of the non-independent viewpoints to be coded, exhaustion and iteration partition method of the CUs of the non-independent viewpoints are omitted in calculation, therefore, the whole CU partition process is accelerated, and efficiency is improved.