In this study we simulated the effect of reconstructing computed tomography (CT) images with different reconstruction kernels by employing a convolutional neural network (CNN) to map images produced by a fixed input kernel to images produced by different kernels. The CNN input images consisted of thin slices (0.6 mm) reconstructed with a sharpest kernel possible on the CT scanner. The network was trained using supervised learning to produce output images that simulate medium, medium-sharp, and sharp kernels. Performance was evaluated by comparing the simulated images to actual reconstructions performed on a reserved set of patient data. We found that the CNN simulated the effect of switching reconstruction kernels to a high level of accuracy, and in only a small fraction of the time that it takes to perform a full reconstruction. This application can potentially be used to streamline and simplify the clinical workflow for storing, viewing, and reconstructing CT images.