The existing panoramic image matching methods are difficult to overcome instability of the corresponding points due to the uneven frequency of projection transformation. In order to improve the robustness and reliability of the corresponding points matching between stereo panoramic image pairs, a feature matching screening method based on panoramic disparity is proposed. First, the concept of panoramic disparity is proposed. Then the parallax value is derived according to the difference of angle. The matching selection is realized through the geometric relation of the core lines of the corresponding points. Experiments show that the proposed method can obtain more stable matching results than the existing algorithms, and the accuracy of corresponding point search and matching is improved.