Measurement of the weld pool surface is a difficult but urgent task in the welding community. It plays an important role in developing the next generation intelligent welding machines, in particular, controlling the weld joint penetration in automatic welding. In this work, the images of front-side free surface and back-side width of the weld pool are synchronously captured with two CCD cameras in tungsten inert gas (TIG) arc welding process based on laser vision, and the three dimensional weld pool surface is reconstructed using designed algorithm. Then, the correlation between the weld pool surface under different weld joint penetration and its back-side width is analyzed qualitatively. The variation of reflected laser dots shape is reversely simulated under different depths of weld pool surface, which represents the different weld joint penetration, after established the standard model of weld pool, incident, reflected ray and imaging plan's mathematical model. It is found that the change of weld pool surface obtained by this method has an intimate immanent correlation with the weld joint penetration. When the weld joint penetration changes from partial penetration to complete joint penetration, the weld pool surface's shape changes from convexity to concave, and its depth is increasing with the increased back-side width. The row curvature of reflected laser dots is diminishing. When the penetration achieves excessive penetration, the depth of the weld pool surface is more rapidly increased, and the row curvature is also increased. The gathered laser dots is drawn back again at parallel direction. The simulated results associate with the measurement very well. The variations of reflected laser dots shape can represent the degree of weld joint penetration. This research lays a good foundation for the control of weld joint penetration using the characteristics of weld pool surface.