Skull gender identification is a method widely used in forensic medicine and anthropology to determine the biological sex of an individual by studying the characteristics of the human skull. However, it is challenging due to the difficulty in obtaining 3D skull data, along with the complex topology and subtle geometric differences of 3D skulls. In this paper, we propose a morphology-guided 3D skull gender identification method by a deep learning framework with an attention mechanism. Based on the skull’s morphological characteristics, the boundary points are firstly extracted as sampled features. By employing the Bidirectional Encoder Representations from Transformers (BERT) method, we embed the skull point cloud boundaries into multiple local patches and use a discrete variational autoencoder to encode point tokens. To enhance the feature learning ability, we apply a masked modeling strategy to point cloud feature inference using a transformer encoder with a self-attention mechanism and introduce a group convolution operation for gender identification. Our method combines deep learning methods with prior knowledge of the skull to achieve good performance on a small dataset. The experimental results demonstrate that our method is effective for 3D skull gender identification.