When manually calibrating water quality models, considerable time and attention are required. Hence, developing an automated model that allows for efficient and objective automatic calibration is highly desirable. The QUAL2Kw model calibrates the QUAL2K automatically using a genetic algorithm (GA). This study analyzes auto-calibration results and selects the optimal criterion for each objective function from six performance criteria. Additionally, a multi-objective auto-calibration was conducted using two kinds of performance statistics as the objective function of the GA. The auto-calibration model was applied to the Yeongsan River and the total maximum daily load (TMDL) was established to achieve water quality goals at specific target points of this river. Among the six auto-calibration results based on a single performance criterion, Nash-Sutcliffe model efficiency (NSE) was the best criterion for calculating fitness through auto-calibration. To consider the calibration accuracies of the TMDL target points and the entire river simultaneously, an objective function using multiple performance criteria, specifically the weighted average of the normalized root mean squares error (CV(RMSE)) and the ratio of the RMSE to the standard deviation of the observed data (the RSR), was selected as the final auto-calibration of the model. The model calibration performance was good across the whole region as well as at the target points.