Abstract: The object of this work is to be able to predict the changes, which occur in the microstructure of metal and alloys during thermomechanical process like rolling. At present reliable model exists for its determination. Optimization of such processes normally demands a combination of several experiments and expensive trials. The final microstructure is dependent on various parameters such as alloy composition of metal, working temperature, local compression rate etc. Determination of grain size of an image of microstructure is difficult to predict with traditional micro-mechanical models. Neural networks however, are ideally suited to such non-linear, multi-parameter problems. In the present work, an attempt has been made to investigate and develop suitable neural network architecture, implementing multi-layer error-backpropagation algorithm, which is appropriate for this metallurgical application. The project lies at the boundary of the practical industrial problems and academic information analysis theory.