The objective of the study was to present the architect, training system consideration and performance of Levenberg-Marquardt back propagation algorithm for manufacturing operation selection that can be associated with STEP-NC file generation. Here STEP-based product data modeling, EXPRESS representation combined with neural network algorithm using feature based design attributes such as dimension ratio, tolerance and surface finish etc. It describes the details of training algorithm steps, training data pattern and performance output of the system in association with object oriented data model in process plan data generation. It pves required criterion for extending the system for machine tools and cutting tools selection with reference to turn-mill process planning requirements. Finally summarizes on importance and methods of binding STEP representation and intelligent Computer Aided Process Planning (CAPP) system development. © 2013 Asian Network for Scientific Information.