The main difficulties of on-line quality control are the availability of on-line product quality measurements. Soft-sensing techniques supply attractive and efficient methods to deal with these difficulties. Soft sensors refer to the modelling approaches to estimating hard-to-measure process variables (e.g. quality variables) from other easy-to-measure variables (e.g. temperature, pressure and flowrate measurements). At present, much more research is concerned with multi-input single-output (MISO) systems than with MIMO systems in the field of soft-sensing modelling. In this paper, some MIMO soft-sensing techniques are studied for estimating multiple product quality variables simultaneously in a hydrocracking fractionator. RBF and fuzzy ARTMAP networks are used to build the models and the latter is shown to be more suitable for MIMO soft-sensing modelling. The issues of data pretreatment and on-line correction, which are very important for the industrial implementation of MIMO soft sensors, are discussed in detail. A useful method using a multivariable fuzzy PID (MFPID) on-line correction algorithm is proposed for the MIMO soft sensors enabling them to adapt with the fluctuation of process operating conditions and uncertain system disturbances. The real application results show that the proposed methods are effective for MIMO soft-sensing modelling and have great promise in industrial process applications.