The paper deals with the design and implementation of an intelligent modular information system (IMIS) for modeling and predictive decision making supervisory control of some important critical processes in a nuclear power plant (nuclear reactor) using selected soft computing methods. The developed IMIS enables monitoring critical states, safety impact analysis and prediction of dangerous situations. It also recommends the operator possibilities how to proceed to ensure safety of operations and humans and environment. The proposed complex IMIS has been tested on real data from a nuclear power plant process primarily used as supervisory information for decision making support and management of critical processes. The core of the proposed IMIS is a general nonlinear neural network mathematical model. For prediction of selected process variables an artificial neural network of multilayer perceptron type (MLP) has been used. The effective Levenberg-Marquardt method was used to train the MLP network. Testing and verification of the neural prediction model were carried out on real operating data measurements obtained from the NPP Jaslovske Bohunice.