A performance recognition method for antimony flotation process is proposed, where mutual coupling and obvious importance difference exist in froth image features. With the combination of the sensitivity index and the principal component analysis, key froth features are acquired. The correlation degrees between key froth features and each preset performance are calculated by using the correlation function, and the synthetic weight value of each feature is generated on the basis of integrating the objective-dynamic weight and subjective-static weight through game theory. Then synthetic correlation degrees between unclassified performance and preset performance are calculated. The industrial on-site experiment verifies the effectiveness of the proposed method which can automatically accomplish on-line monitoring of the flotation performance.