Abstract: Given the complexity of the wastewater treatment process and the difficulties in online instrument maintenance, soft measurement with the use of a computer has become a valid way to evaluate the performance of the wastewater treatment process.We propose a novel online soft measuring model based on multi-attribute Gaussian kernel function FAST relevance vector machine (MAG-FASTRVM).This novel model establishes a Bayesian matrix with MAG kernel functions and accelerates the update speed with fast marginal likelihood algorithm.Experiment results verify that the proposed model can reduce the number of relevance vectors and improve prediction accuracy compared with the support vector machine, the relevance vector machine, the FASTRVM, and several multi-kernel function FASTRVMs.The computation time of modeling is also significantly reduced.The proposed model is effective for online soft measurement in the wastewater treatment process.