In the typical urban environment, GPS standalone can hardly satisfy the requirement of continuous positioning, particularly in the weak signal or no signal environment. Fortunately, this problem can be solved by using the integration method based on GPS and non-LOS (Line of Sight) positioning such as RFID positioning. To solve the problem of computational load in the traditional GPS/RFID integration based on all selection of RFID observations, this research proposed an optimal selection method of RFID observations with the minimum positioning error under the weighted observation condition. The experiment results show that the geometric distribution of precision in the environments, such as shadow or blockage, can be improved by using the proposed method, and both availability and accuracy of positioning can be improved efficiently. Moreover, the computational load can be decreased dramatically through this proposed method.