At present, the agriculture images are increasing greatly day by day. It becomes increasingly difficult for users to find out the desired images. Building the database to store images and realizing the automatic retrieval of images is very important in agriculture field. Based on the study of visual feature retrieval technology of images, this article puts forward a new retrieval algorithm using color, texture and shape features according to the characteristics of crop image database. The retrieval accuracy rate has been improved greatly and by reasonably setting the feature library, the retrieval efficiency is greatly improved. An experimental system is constructed through programming. We establish a relatively complete test database for crop images and generic image database, demonstrate the effectiveness of the proposed retrieval algorithm. Finally, the retrieval performance using different features is evaluated and compared, and the problems and future works are analyzed.