This paper desriibes a common recognition probem, related to an industrial enviroment which has a determined number a different ojects that are needed to be recognized. the manufacturing enviroment is charaterized by rapid change, origanting new challenges and problems to the producation and operation manager in the industry. in response to the need for fast and flexible manucfaturing incresing attention is being given to integration of computing techologies with the manufacturing systems leading to the development of fast and flexible manufacturing systems aided with high performance vision capabilities. in such difficult environments, where objects to be recognized can be dirty and illumination donditions cannot be sufficiently controlled, the required accuracy and rigidity of the system are cristical features. our approach and proposal is based on neural netwroks. the system works with bi-demensional imagees of the object which are processed briefly before the recognition step. the purpose of the system is to reconize air-coditioning objects for avoiding erroneous identifications due to a large variety of size and kinds of objects. experimental results on inspection and recognation of a large variety of air-conditioning objects are provided to show the performance of the different of the differnt network architectures studied.