As a fundamental machine learning method, the K-Nearest Neighbor (KNN) algorithm is widely used in fields such as medical diagnosis and financial risk management due to its simplicity and interpretability. However, the increasing scale of data and prevalence of multi-party collaboration have raised significant privacy concerns due to the potential leakage of sensitive data during KNN computation. While existing Secure Multi-Party Computation (SMPC) schemes offer privacy protection, they often suffer from high computational and communication overhead. In this paper, we construct an efficient and secure KNN query scheme (ESKNN) based on three-party computation. To support ESKNN, We propose a three-party efficient and secure computation framework (ES3PC) based on a non-uniform bit-width design, which reduces computational and communication overhead. Compared to the existing works we have known, our scheme not only decreases communication overhead but also runs 48.8 \(\times \) faster than the state-of-the-art approach. Finally, we provide complexity analysis, formal security proofs and experimental evaluation to demonstrate the correctness and efficiency of the proposed protocols.