An automatic technique to improve the surface model of the shoe-last bottom via Bezier networks is presented. In this technique, the surface model is generated via Bezier networks based on surface points, which are measured via line projection. The measurement procedure is performed by a Bezier network based on the line position. Thus, the model provides a continuous surface pattern of shoe-last bottom with high accuracy because the model passes through all control points of the physical surface. Furthermore, the network reduces the operations and memory size to calculate the surface. It is because the computational model is implemented with less mathematical terms than the traditional models. Additionally, the network and the laser line compute the vision parameters to avoid external calibration, which increases the inaccuracy of the surface model. Also, the model adapts the surface of the shoe-last bottom to the plantar surface. Thus, the proposed technique improves the accuracy, speed, and memory size of the traditional surface models. This improvement is proven by an evaluation based on the traditional surface models, which perform the model of the shoe-last bottom. The evaluation shows experimental results, which provide evidences of the contribution of the proposed technique.