Multilayer feedforward neural networks with error backpropagation algorithm have been applied to analyze the drawing process of PET films. Six kinds of process conditions, four kinds of material structures and six kinds of material properties were used as inputs and outputs chosen to analyze the relationships between process conditions, material structures and properties. It was shown that the trained neural networks represent a small error between the target and the corresponding estimated values. It has been shown quantitatively that material studies play an intermediate role in the relationship between process conditions, material structures and properties. It was also shown that model-free simulator predictions by neural networks are more reliable than the corresponding conventional non-linear regression analysis in the complex polymer processing.