Inaccurate extrinsic parameters are one of the major errors of a mobile LiDAR system (MLS). As different types of sensors with incommensurable precisions are integrated in an MLS, the extrinsic parameters cannot be easily isolated and estimated, especially in an automatic mode. To address this issue, this paper proposes an automatic extrinsic self-calibration method for an MLS based on planar and spherical features. First, the planar and spherical features are automatically extracted from scanned point cloud data of different strips using the Random Sample Consensus algorithm, and the corresponding features are matched after the extraction. Secondly, a rigorous relationship is established between the direct geo-referencing equation, the extrinsic parameters, and the geometric constraint model. Thirdly, the extrinsic parameters are calibrated by minimizing the sum of the squares of distances from points on the feature surface to the matched reference feature. Four datasets collected by using four types of MLS were used to verify the proposed method. As a result of combining two geometric features into a single self-calibration adjustment, the experimental results show that the proposed method is superior to the conventional plane-based method in terms of the positional accuracy. The standard deviation of the distance between the check features of the four datasets collected by several mobile platforms before (after) extrinsic calibration were 0.051 m (0.024 m), 0.060 m (0.018 m), 0.029 m (0.009 m), and 0.354 m (0.070 m), which demonstrated the high compatibility and practicality of the proposed method.