Summary: Feature recognition is considered a vital link between Computer-Aided Design (CAD) and Computer-Aided Process Planning (CAPP) or the first step towards the integration of CAD, CAPP, and CAM. However, most known feature recognition systems do not have a capability to recognise new features. In this paper, we propose a new methodology that works with the mechanism of the neocognitron, i.e., neural network modelling for pattern recognition. The attributed adjacency graph (AAG) information extracted from a B-Rep solid model is converted to attributed adjacency matrices (AAM) that can be applied as input data for the recogniser to train and recognise form feature patterns. With this technique, the system can self-reconstruct its recognition abilities for new features by learning without a priori knowledge and software modifications. This system is developed as one of the function models of a generic CAPP model which can provide users with a more generic and intelligent interface for the integration of CAD and CAPP.