Abstract: Finger phalangeal prints are stable and unique for each individual, thus can be used in personal authentication. A novel personal recognition method is proposed based on the finger phalangeal print in this paper. The captured image is first preprocessed to segment out fingers and rotate them to the horizontal direction; Sobel operator is used to get the horizontal gradient which is binarized and projected to the horizontal axis. The 1-D projection is then denoised with wavelet and downsampled to get the feature vector. Finally, the similarity of the two matching vectors is measured by using cosine function. The proposed method was tested on the database which contains 1900 samples from 190 different palms. The equal error rate (EER) was no more than 0.67% and one match time consumption was less than 2ms. Experiments suggested that a high recognition accuracy and efficient matching performance can be achieved with the proposed algorithm, which can meet the requirement of real time searching in large palm database.