A Dissolved Air Flotation System was optimized using factorial experimental design at different turbidity conditions. Using cluster analysis, historical water quality data was classified into low (0 – 181 NTU) and high (182 – 323 NTU) turbidity cluster. A multiple regression analysis was performed for each of the cluster which revealed that for low turbidity cluster eight factors are significant while only three factors were significant for the high turbidity cluster. Using this knowledge, a factorial experiment was performed for both clusters. A model was generated with R2 adjusted = 86.25% for low turbidity cluster and R2 adjusted = 67.34% for the high turbidity cluster. The accuracy of the models was tested on actual conditions and showed 92.60% accuracy for the low turbidity cluster and 88.12% accuracy for the high turbidity cluster.