X-ray fluorescence analyzer is in deficiency and manual detection results in large time delay in present online detection of mineral flotation. Aiming at these issues, a novel optimal setting of reagent dosage based on least squares support vector machine (LS-SVM) is proposed. Through obtaining a large amount of history process data from flotation running, LS-SVM is used to implement optimal setting of reagent dosage for flotation process. The simulation results based on experiment data prove the effectiveness of the proposed method.