Provided is a load sharing method, including that public network addresses are configured in a form of address pools; after traffic transmitted by user devices is received, the user devices are grouped according to traffic models transmitted by the user devices; CPU allocate the public network addresses to the user devices and translate private network addresses of the user devices into the public network addresses, wherein each of the address pools includes a plurality of address blocks, each of the plurality of address blocks corresponds to one CPU, each user group includes a plurality of user blocks, and each of the plurality of user blocks corresponds to one CPU. A load sharing apparatus is also provided. Thus, uneven sharing of public network address resources among a plurality of CPUs and loads of user devices can be avoided, thereby not only improving the processing speed of the CPUs for address translation, but also obtaining an excellent load sharing effect.