Abstract: A kind of MIMO soft-sensing modeling is studied in this paper. In an oil refinery,hydrocracking fractionator produces three products,light naphtha,heavy naphtha and jet fuel,and the endpoints of these three products are the main specification for evaluating the product quality.Because the endpoint is a kind of analysis indicator.it is very difficult to obtain the value online.Fuzzy ARTMAP network is used to build the MIMO soft-sensing models based on practical process data for estimating three endpoints of hydrocracking fractionator online.The generalization results can satisfy the precision reauiremnnts of process technology.It proves that fuzzy ARTMAP has the ability of classification and multidimensional nonlinear maps Considcring the frequent changes of the practical operating conditions,the soft sensor should have the adaptive ability to overcome the uncertain system disturbances and the fluctuation of the operating conditions.A multivariable fuzzy PID online rectification algorithm is then proposed for this objective.Combining with the proposed online rectificati on algorithm,the fuzzy ARTMAP-based soft sensor provides an effective method for multivariable nonlinear MIMO soft sensing modeling.