We present a data-driven method for automatically generating a 3D cartoon of a real 3D face. Given a sparse set of 3D real faces and their corresponding cartoon faces modeled by an artist, our method models the face in each subspace as the deformation of its nearby exemplars and learn a mapping between the deformations defined by the real faces and their cartoon counterparts. To reduce the exemplars needed for learning, we regress a collection of linear mappings defined locally in both face geometry and identity spaces and develop a progressive scheme for users to gradually add new exemplars for training. At runtime, our method first finds the nearby exemplars of an input real face and then constructs the result cartoon face from the corresponding cartoon faces of the nearby real face exemplars and the local deformations mapped from the real face subspace. Our method greatly simplifies the cartoon generation process by learning artistic styles from a sparse set of exemplars. We validate the efficiency and effectiveness of our method by applying it to faces of different facial features. Results demonstrate that our method not only preserves the artistic style of the exemplars, but also keeps the unique facial geometric features of different identities.