This paper proposes a novel point-based representation called Gaussian surfels for high-quality surface reconstruction. Gaussian surfels combine the advantages of 3D Gaussian points and surfels by setting the z-scale of 3D Gaussian points to zero, effectively flattening the 3D ellipsoid into a 2D ellipse. This design provides clear guidance to the optimizer, treating the local z-axis as the normal direction, which improves optimization stability and surface alignment. A self-supervised normal-depth consistency loss is introduced to address the issue of zero derivatives in the local z-axis. Monocular normal priors and foreground masks are incorporated to enhance reconstruction quality, mitigating issues related to highlights and background. A volumetric cutting method is proposed to remove erroneous points in depth maps generated by alpha blending. Finally, a screened Poisson reconstruction method is applied to the fused depth maps to extract the surface mesh. Experimental results show that the proposed method achieves superior performance in surface reconstruction compared to state-of-the-art neural volume rendering and point-based rendering methods. The method demonstrates a good balance between reconstruction quality and training speed. The paper also presents ablation studies showing the effectiveness of various loss terms in improving reconstruction quality. The method is evaluated on the DTU and BlendedMVS datasets, showing significant improvements in surface accuracy and rendering fidelity. The results demonstrate that Gaussian surfels can achieve high-quality surface reconstruction with a good balance between computational cost and reconstruction quality.This paper proposes a novel point-based representation called Gaussian surfels for high-quality surface reconstruction. Gaussian surfels combine the advantages of 3D Gaussian points and surfels by setting the z-scale of 3D Gaussian points to zero, effectively flattening the 3D ellipsoid into a 2D ellipse. This design provides clear guidance to the optimizer, treating the local z-axis as the normal direction, which improves optimization stability and surface alignment. A self-supervised normal-depth consistency loss is introduced to address the issue of zero derivatives in the local z-axis. Monocular normal priors and foreground masks are incorporated to enhance reconstruction quality, mitigating issues related to highlights and background. A volumetric cutting method is proposed to remove erroneous points in depth maps generated by alpha blending. Finally, a screened Poisson reconstruction method is applied to the fused depth maps to extract the surface mesh. Experimental results show that the proposed method achieves superior performance in surface reconstruction compared to state-of-the-art neural volume rendering and point-based rendering methods. The method demonstrates a good balance between reconstruction quality and training speed. The paper also presents ablation studies showing the effectiveness of various loss terms in improving reconstruction quality. The method is evaluated on the DTU and BlendedMVS datasets, showing significant improvements in surface accuracy and rendering fidelity. The results demonstrate that Gaussian surfels can achieve high-quality surface reconstruction with a good balance between computational cost and reconstruction quality.