Artificial neural network model is developed for the prediction of phase transformation of steel from austenite, and thus construction of the continuous cooling transformation (CCT) diagram. The model for prediction of transformation temperatures from steel composition is developed using data from published CCT diagrams. The trained network sometimes fails to predict the sequence of the phase transformation, contradicting the fundamentals of metallurgy. To avoid such limitations of data driven models and to make the models truthful and reasonable from metallurgical standpoint, prior knowledge is incorporated using genetic algorithm, through modifying the weights and biases of a trained neural network. The conventionally backpropagated multi-layered perceptron is modified from error minimization as well as knowledge incorporation point of view through formulation of the problem in both single and multi-objective optimization domains. The predictions of six transformation temperatures by the new models are found to be significantly better than the conventionally trained model.