There are more than one measure devices during the craft launching proceeding, reliable craft orbit measurement selection is important for launching monitor and safety control. This paper proposes a novel orbit selection method by knowledge based neural network, which combines the human knowledge and orbit samples based training together. The false spur orbit data are deleted according the data continuity rule at first, then the difference between orbit data is computed, and at last a neural network is trained to test reliability of each data and select the orbit with high precision among all orbit data. Simulations and experiments show the efficiency and reliability of our method.