The biggest challenge in channel assignment for dense, multi-cell/AP wireless LANs is to arrange co-channel cells so as to maximize the aggregate network throughput. Most previous work models this problem as a vertex coloring problem. In this paper we model it as a non-linear optimization problem to maximize overall network throughput. We prove that the new optimization problem is NP-hard and vertex-coloring is a simplified case. We then propose a polynomial time heuristic algorithm called MIF (Most-Interfered-First) for channel assignment. The performance for a line topology is analyzed. Simulations for random topologies show that MIF consistently produces better network throughput than vertex-coloring based heuristic algorithms with less computation cost.