The design of a real-time detection system for agriculture field pests was developed. The collecting-conveying device could collect and adjust the posture of pests automatically, and send them into the area of CCD camera with proper speed. The chamber and collection system could produce even and invariable light, and collect series of images information at real time. The self-adaptive enhancement techniques were used to improve the images of agriculture field pests sample, the pests image can be separated from the background by using the auto-threshold method. The characters of a pest, such as perimeter, invariant quadrature, and so on, were taken as its characteristics to identify nine types of pests by using a classifier based on BP NN. The experiment shows that the system is practical and feasible.