While the first part of this paper was concerned primarily with the issues of system structure and associated learning control laws, the second part presents a methodology for constructing the rule-base from learned data. The approach, based on a simplified fuzzy control model, is systematic, simple but efficient. The proposed system is applied to the problem of multivariable control of blood pressure, which is characterized by strong interactions and pure time delays in controls. It is shown that the effects of loop interaction are removed automatically by the learning scheme so that a decoupled control structure can be built. Moreover, the problem of pure part delay is easily tackled due to the iterative property. By defining some performance measures such as AD, RD, FRD and IP, the behaviour of the proposed system in terms of the learning ability (adaptability), reproducibility and robustness are evaluated through a number of simulation studies. Some important conclusions are drawn from these studies.