The component concentrations in sodium aluminate solution are very important in the process of alumina production, they represents the product quality. At present they can not be measured online, so the optimal operation is hardly to be achieved. To deal with this problem and based on the character of process industry data, we propose a RKPLS (Robust Kernel Partial Least Squares) soft sensing method by combining robust algorithm and kernel transformation to predict the component concentrations in sodium aluminate solution. Industry experiments are conducted in the alumina production process and the results show the effectiveness of this method.