In the development of an object pattern recognition system, feature construction is always the problem issue. Due to the large amount of information contained in three dimensional (3D) objects, features extracted to efficiently and sufficiently represent 3D objects are difficult to obtain. Thus, current commercially available object recognition systems mostly emphasize the classification of two dimensional objects or patterns. This work presents a paradigm to develop a complete 3D object recognition system that uses simple and efficient features, and supports the integration of CAD/CAM models;In this research, several proposed algorithm for extracting features representing 3D objects are constructed based on the properties of the Radon transform. Two of these algorithms have been successfully implemented for manufacturing applications. The implemented systems use the artificial neural network as the classifier to learn features and to identify 3D objects. A statistical model has also been established based on the output node values of a perceptron neural network to predict the future misclassifications of features which have not been learned by the neural network in the training stage.