This study aims at real-time measurement /reconstruction of the specular three-dimensional (3D) weld pool surface under strong arc in gas tungsten arc welding (GTAW) using an innovative machine vision system. A dot-matrix laser pattern is projected on the specular weld pool surface. Its reflection from the specular surface is intercepted at a distance from the arc by a diffuse plane. The intercepted laser illuminates this plane producing an image showing the reflection pattern. The deformation of this reflection pattern from the projected pattern can be used to derive the 3D shape of the reflection surface, i.e., the specular weld pool surface. The image of the reflection laser pattern is captured by a separate camera. To measure and reconstruct the weld pool surface in real-time, a fast solution is needed to derive/reconstruct the weld pool surface from the extracted laser reflection pattern and the known incident laser pattern. Based on careful analysis, the underlying reconstruction problem is formulated mathematically. An analytic solution has been proposed to solve this formulated problem resulting in that the surface reconstructed in average 3.04 ms such that a weld pool surface can be measured/reconstructed at 30Hz from image acquisition, processing to surface reconstruction. A real-time system is thus established to measure and reconstruct the weld pool surface in GTAW in real-time during welding. Simulations and real-time experiments verified the effectiveness of the established system.