Abstract: Under the influence of palm pose, illumination variation and complicated backgrounds; contact-less hand image recognition systems have unsatisfied recognition effect. To solve these question, a contact-less hand image recognition method based on convolutional neural network (CNN) was proposed. First the hand image was segmented from the background with a skin-color thresholding method.A finger base point detection method was used to find the finger base points. These points could be used to locate region of interest (ROI) of palm. Then CNN was used to extract hand features which could be basic for classification. The presented method was tested on publication hand image database. The results demonstrate that proposed algorithm can be used to improve the recognition accuracy.