People are paying more attention to the use of Spherical Panorama Images (SPIs) for many applications. In order to apply SPIs in photogrammetric application such as land mapping or navigation like frame images do, conjugate points matching and the relative relationship between SPIs are important issues. Through observing the moving pattern of conjugate points, the relative positions and orientation between camera stations may be solved. In this study, images captured by Ladybug 5 system were used for experiment, image features were extracted and matched by Speed-Up Robust Features (SURF) algorithm (Bay, 2008), and the concept of Random Sample Consensus (RANSAC) was applied to improve the accuracy of conjugate points matching. Although RANSAC general model is not well enough to detect the features on SPIs, we proposed a method using the Essential Matrix model to improve this deficiency. Once the conjugate points are found, the relationship between image stations can be explained by Essential matrix, the rotation and translation parts can be extracted up to scale. Similar to that of frame camera, four possible solutions can be found, the angle between two image stations is used to judge the correct solution. The results show that the quantity and quality of corresponding pairs influence the accuracy of the relative positions and orientations between two images. Although the error matching pair can be found and removed by RANSAC, the distortion comes with projection still make trouble for SURF algorithm. A suitable way is apply image matching on the spherical space to improve the quality of corresponding pairs.