We present a fast GPU-based method for rendering boundary sur- faces from dynamic 3D particle clouds. Instead of extracting and tracking polygo- nal meshes of boundary surfaces in 3D world space, we devise a new meshless 2D screen-space approach. First, we generate a smooth view-dependent depth map from the 3D particle cloud. Then, surface normals are derived directly from the depth map and lighting calculations are carried out in screen-space. The advantage of our method over existing rendering approaches for particle-based objects is the fact that we do not have to extract 3D or 2D meshes at any time. We focus on free surfaces of 3D particle-based, interactive liquid simulations. Thus, we extend the basic algorithm with efficient reflection and refraction approximations.