Existing methods for 3D local feature description often struggle to achieve a good balance in distinctiveness, robustness, and computational efficiency. To address this challenge, a novel 3D local feature descriptor named Histograms of Projected Normal Vector Distribution (HPNVD) is proposed. The HPNVD descriptor consists of two main components. First, a Local Reference Frame (LRF) is constructed based on the covariance matrix and neighborhood projection to achieve invariance to rigid transformations. Then, the local surface normals are projected onto three coordinate planes within the LRF, which allows for effective encoding of the local shape information. The projection plane is further divided into multiple regions, and a histogram is computed for each plane to generate the final HPNVD descriptor. Experimental results demonstrate that the proposed HPNVD descriptor achieves a good balance among distinctiveness, robustness, and computational efficiency. Moreover, the HPNVD-based point cloud registration algorithm shows excellent performance, further validating the effectiveness of the descriptor.