We present a complete solution to evaluate lung tumor motion from planning CT scan, patient positioning for treatment, to treatment delivery, without the requirement of implanted fiducial markers. Tumor motion is evaluated at three stages of the treatment process. After 4D CT scans for treatment planning, 3D images are reconstructed for 10 phases. Tumors are contoured at one phase. A rigid 3D registration is applied to register the contoured tumor to other phases, so that a tumor motion pattern can be obtained. Tumor motion prior to each treatment is evaluated by locating tumors directly on every raw projection of daily localization CBCT scans. In order to remove the strong signals from under- and over-lying anatomic structures, a novel subtraction procedure is performed using DRRs generated from the planning CT and contours. Finally, tumor motion during treatment is evaluated by analyzing MV treatment beam images. Every frame of image is evaluated in three components, with separate DRRs generated from planning CT: all non-moving structures, the tumor, and other moving structures. The locations of three components are determined though a registration process. This technique has been evaluated in phantom studies, with an overall accuracy within 1 mm. It also has been applied to lung cancer patients undergoing radiotherapy. To quantify the clinical impact, the time in which the tumor excursion exceeded the planning margin was evaluated. This technique is capable of accurately characterizing lung tumor motion on a daily basis, without implanting fiducial markers and without additional imaging dose. It provides daily verification that the tumor range is within prior estimates and covered by the treatment planning volume.