An adaptive control in the robotic GMA welding is employed to monitor information about weld characteristics and process parameters as well as to modify those parameters to hold weld quality within acceptable limits. Typical characteristics are the bead geometry, composition, microstructure, appearance, and process parameters which govern the quality of the final weld. The main objectives of this thesis are to realize the mapping characteristics of bead width through learning. After learning, the neural estimation can estimate the bead width desired form the learning mapping characteristic. The design parameters of the neural network estimator(the number of hidden layers and the number of nodes in a layer) are chosen from an estimation error analysis. A series of bead of bead-on-plate GMA welding experiments was carried out in order to verify the performance of the neural network estimator. The experimental results show that the proposed neural network estimator can predict the bead width with reasonable accuracy and guarantee the uniform weld quality.