Abstract: Palmprint images contain rich unique features for reliable human identification, which makes it become a very competitive topic in biometric research. From a low resolution palmprint image, the information of principal lines and wrinkles can be obtained to realize palmprint recognition. The direction feature of palmprint lines is an effective feature. But how to effectively fuse the direction feature and other palmprint line features is an open problem in palmprint recognition. In order to solve the problem, a palmprint recognition algorithm based on palmprint field features is proposed in the paper. In the method, data fields and wavelet packed entropy is used to construct palmprint data field and extract a new palmprint feature, the palmprint field feature. The field feature is the combination of the structural feature and direction feature. First, the data field theory is introduced into palmprint recognition field and each point in the palm lines is seen as a data point with unit mass to map an enhanced palmprint image from gray space to the corresponding potential space. In the space, all of points in the palm lines will be affected by other points to form a palmprint image data field. Because the distribution of palmprint data field is affected by the thickness, direction and distribution density of palm-lines, a wealth of the structural and direction information of palm-lines are provided by palmprint data field. For the sake of improving the distinguish ability, the palmprint data field data is decomposed into a relative palmprint data field, an absolute palmprint data field and a direction data field. The absolute data field can make a rough distinction between background and targets, the edge information is highlighted in the relative data field and the direction information of points in each palm-line is obtained in the direction data field. Next, the different sub-data fields are decomposed by wavelet packet transform and the entropies for all nodes for wavelet packet are calculated. These wavelet packed entropies can represent the features of energy distribution of different sub-data fields in different nodes. Finally, all of features of each sub-field are joined into one palmprint field feature, which is fed to backpropagation neural networks for classification. The experimental results illustrate the effectiveness of the method.