On-board cone-beam computed tomography (CBCT) provides volumetric information of a patient at treatment position and improves the setup accuracy and may be used for dose reconstruction. However, there is a great concern that repeated use of CBCT during a treatment course delivers too much extra dose to the patient. To reduce the CBCT dose, one needs to lower the total mAs of the x-ray current, which usually leads to reduced image quality. Our goal of this work is to develop an effective method that enables to achieve clinically acceptable CBCT image with as low as possible mAs without compromising quality. An iterative image reconstruction algorithm based on a penalized weighted least-squares (PWLS) principle was developed for this purpose. To preserve edges in the reconstructed images, we designed an anisotropic penalty term of a quadratic form. The algorithm was evaluated with a CT quality assurance phantom and an anthropomorphic head phantom. Compared with conventional isotropic penalty, the PWLS image reconstruction algorithm with anisotropic penalty shows better resolution preservation.