Distribution network planning is generally a high dimension and nonlinear combinatorial optimization problem. To cope with the computational difficulty of this problem， an ordinal optimization theory based algorithm for solving the optimal solution is proposed， which can best satisfy the engineering requirements. However complicated the objective function and heavy the calculating workload， and however complex the subset of solution， the ordinal optimization theory based algorithm is sure to attain a good enough solution with high probability. Furthermore， the algorithm proposed is suitable for distribution system planning whose objective function is the total cost comprising the fixed investments， maintenance costs and feeder resistive losses. It has been found from the planning results of an IEEE distribution system with 4 substations and 61 feeders are superior to those of the genetic algorithm， demonstrating the feasibility and effectiveness of the approach proposed.