While gas metal arc welding of aluminum has been widely implemented in automotive industry, it is prone to process disturbances. Therefore, an effective monitoring method is necessary to ensure the weld quality (e.g., weld bead height, width and penetration). In this study, an online method was developed to estimate the weld bead geometry using the combination of the real time welding signals, characteristics of weld bead geometry and a liquid surface model. The liquid surface model can represent the steady-state geometry of the weld pool with varying characteristics, such as volume, boundary conditions and surface tension. To model the geometry of the weld pool, the weld pool moved and updated its geometry in response to the merging of each new droplet. The detachment time and size of the new droplet were obtained by analyzing real time welding voltage and current. The surface tension was characterized by the effective contact angle measured under various welding conditions. Moreover, the effective contact angles and radius of arc-affected zone were introduced to significantly simplify the complicated thermal behavior during welding. Finally, the weld bead geometry was estimated using real welding signals in both globular and spray transfer modes and validated with the experimental results. This method predicts weld bead geometry with satisfied accurancy for engineering use and in a comparable speed as the welding speed; and thus can be applied as an online monitoring method for the weld bead quality.