This paper presents a sectored antenna array indoor positioning system (IPS) with neural network (NN) technique. The hexagonal positioning station is composed of six printed-circuit board Yagi-Uda antennas and Zigbee modules. The values of received signal strength (RSS) sensed by wireless sensors were used to be the information for object’s position estimation. Two NN models, including NN with back-propagation (BP) learning algorithm and probabilistic NN (PNN), were applied to perform the positioning work for a comparison. In the experiments, an 8x8 square meters indoor scene was performed and 288 points and 440 points were experimented in this area. The positioning results show that both NN models have the average error less than 0.7 meter. In other words, the proposed positioning system not only has the high positioning accuracy, but also has the potential in real application.