The cascaded neuron network model is used to recognize 3-D targets. In order to improve the recognizing rate, several methods of choosing the construct and algorithm are proposed. They can be used together or solely according to the requirement. The invariance encoding, the algorithm, the binarizing of the interconnection weights and the selecting methods of the training samples, and so on are studied. The computer simulation of recognizing three plane models is completed based on this cascaded neuron network model. The recognizing rate of the model is over 96% as the three planes arbitrarily positioned and directed.