In the area of engineering design, the birth of a CAD model typically undergoes iterative modifying across three distinct phases: conceptual design, preliminary design, and detailed design, which is a very time-consuming process. Facing the increasing demand of CAD models in various scenarios, to speed up the process, this paper proposes a novel strategy to generate feature-based parametric CAD models based on free-hand sketches, bridging the gap between conceptual design and detailed design. Meanwhile, the strategy considers the context information of features, which constructs logical dependencies between features and hierarchically organizes the features in a parametric CAD model. If users modify one feature, other features adjust accordingly. In Experiments, we establish a dataset to train the proposed networks and verify the efficiency of the proposed method on users’ sketches and public dataset. The code is available at https://github.com/dishy313/sketch2Feature-Net .