The machining accuracy of thin-walled blades significantly impacts the performance of aeroengines. Existing methods for predicting manufacturing deformation of thin-walled blades often focus on the single-stage machining process and single type of error, neglecting the error propagation and coupling of multiple types of errors inherent in multistage processes, which limits their applicability to real-world machining scenarios. Therefore, this paper introduces multitask learning to predict various manufacturing errors for thin-walled blades, and proposes a data-driven multistage machining comprehensive error prediction method. In this method, single-task Gaussian process regression (STGPR) is used to predict the comprehensive error in the pre-finishing milling stage, and multitask Gaussian process regression (MTGPR) is used to predict the comprehensive error after finishing milling. By learning the correlation between different tasks, the MOGPR can improve the prediction accuracy of the contour, position, and torsion errors of thin-walled blades. Additionally, the error propagation regulation can be revealed by the established model. The effectiveness of the proposed method is demonstrated through the compressor blades. The results show that the machining error of the previous stage will significantly increase the contour error of the next stage. This research can provide guidance for the machining error control of thin-walled blades.