This paper presents a new method to reduce halo artifacts in Computed Tomography (CT) images, which blur the image edges and descend the image quality. Off-focal radiation is main ill factor leading to these halo artifacts. To reduce the off-focal effects, we adopt a pre-processing method to filter the sampling data directly before reconstruction, instead of post-processing method such as edge enhancing on the reconstructed images. Different from the prior pre-processing method, we also present a new and more practical method to obtain convolution kernel and a simpler formula in this paper. Results show that after calibration by the new practical method, the off-focal radiation effects were reduced efficiently, the edges were enhanced and the image quality was improved without increasing noise.