With the deepening of oilfield development, the remained oil which shows some effects on oil displacement efficiency is formed in the reservoir pore-throat. At present, the study on the remained oil is almost performed by doing microscopic displacement experiments with the visual glass etching model. Research on remained oil shape in the model can provide important reference for secondary oil recovery and tertiary oil recovery of oilfield. In this work, the geometric feature parameters of remained oil are taken as the input of neural network for classification. Through the training and testing of the BP(Back- Propagation) neural network, good recognition rates can be achieved, which enables fast and accurate classification and identification for remained oil shape.