Direct volume rendering (DVR) is widely used in visualization. There are various DVR methods, such as ray casting, splatting, 2D texture slicing, and 3D texture slicing. The images generated by these methods are somewhat different. Even with the same DVR method, different rendering parameters also produce different images. As the direct volume rendered images will be perceived by human beings, it is important to investigate whether the visible differences between two images will be observed. In this poster, we apply a perceptually-based comparison metric, which is based on Visible Differences Predictor (VDP) developed by Daly, to compare the direct volume rendered images. We also show how to use this metric to predict the alpha threshold value for early ray-termination in a ray-casting algorithm. Experimental results demonstrate that our approach provides an alternative way to evaluate the quality of directed volume rendered images. A new perceptually-based acceleration technique for DVR can be developed based on our work.