Welding and manufacturing process of pipelines is mainly organized with root-pass welding and fill-pass welding. Of them, root-pass welding is very difficult to automatize, compared to normal welding, and it doesn't guarantee high welding quality. Root-pass welding quality can be represented by back-bead width and back-bead height, and there are a variety of studies to predict back-bead geometry, currently. In this study, a variety of welding experiments were carried out to optimize root-pass welding process using GMA process. Based on the experimental results, optimal welding conditions were selected after analyzing correlation between welding parameters and back-bead geometry. Then, effectiveness of empirical models developed was compared and analyzed, and optimized empirical models were finally developed for predicting back-bead by analyzing the main effect of each factor which affects back-bead geometry and their influence on interaction.