As an essential function in Internet routers, address lookup determines overall router performance. The most important performance metric in software-based address lookup is the number of memory accesses since it is directly related to lookup time. This letter proposes an algorithm to perform efficient binary search for IP address lookup. The depth of the proposed binary tree is very close to the minimum bound, and hence it results in much smaller number of worst case memory accesses compared to previous schemes. The proposed algorithm requires comparably small size of memory, and it can be used for software-based address lookup in practical Internet routers.