The flotation process often utilizes multiple features of froth image for identification. However, the existing methods overlook the classification information generated by the interactions among these features of froth image and do not quantify the role of each feature in the classification. Therefore, a weighted naive Bayesian algorithm based on multiple relationship constraints is proposed. First, a multineighborhood radius set is calculated based on the different distributions of froth image features. Then, the multiple relationships of correlation, complementary gain, and interaction gain between features of froth image are defined by considering multineighborhood mutual information and multineighborhood conditional mutual information. Second, according to the proposed objective evaluation function, the features of the froth image are weighted by Naive Bayes (NB), and the flotation condition is finally recognized. Experimental results show that the optimal feature subset selected by this method can effectively identify flotation conditions.