Artificial neural networks (ANN) are used for modeling of industrial processes. However, most of the published papers deal with small or medium scale systems. One of the possible reasons, the slow learning or non convergence of large scale networks can now be overcome by the use of non-random initial connection weight algorithm. The developed ANN process model may be optimized, after the elimination of non-relevant inputs and hidden-layer "neurons". Causal relationships may be extracted from the ANN process model. This paper describes the experience acquired using these algorithms during the last six years in developing ANN models of industrial plants. Examples are given of an activated-sludge urban wastewater treatment plant and of a batch reactor for the production of organic chemicals.