Currently, radiography with C-arm systems is playing a major role in the assessment of arthritis. However, the radiographic two-dimensional projection images of joints often interfere with physicians' efforts to better understand and measure the structure changes of joints due to the overlap of bone structures at different depths. An accurate, low-cost, and practical three-dimensional (3D) reconstruction approach of joints will be beneficial in diagnosing arthritis. Toward this end, a novel method is developed in this paper based on a C-arm system. The idea is to apply the shift-and-add algorithm (commonly used in digital tomosynthesis) on the segmented projection images at multiple angles, which results in accurate reconstruction of the 3D structures of joints. The method provides a new solution to precisely distinguish objects from blurring background. The proposed method has been tested and evaluated on simulated cylinders, a chicken bone phantom with known structure, and an in vivo human index finger. The results are demonstrated and discussed.