The trade-off between accuracy and efficiency is of major concern in rapid prototyping (RP), where STL is the most commonly used format to present a CAD model. A STL model needs to be sliced into layered contours before it is used to build a RP object. The challenge to achieve the accuracy/efficiency trade-off becomes critical when a STL model contains a large number of triangles. Previous work by Zeng et al. [19] proposed an efficient slicing method that speeded up the layer decomposition of the STL model by adopting a one-dimensional LDNI sampling, which is independent of geometric complexity. However, the accuracy of the generated contours is not entirely satisfactory for the model with sharp features, even by adopting a high sampling resolution. Therefore, the objective of this work focuses on achieving accurate layer contours in a more effective manner without scarifying the high efficiency. In this paper, a grid-structured point representation with neighborhood information embedded, Surfel-Grid, is proposed, which consists of a Grid (formed by two in-plane orthogonal rays), and surfels (sampling points with normal vectors) located on the edges of the Grid. After discretization of a STL mesh model into layers of Surfel-Grid, loop construction is then applied to each of the Surfel-Grid layer. The accuracy of resulting layered contours is improved due to the adoption of two dimensional sampling and the generation of new reconstructed features. Taking advantage of the neighborhood information and cell validity of Surfel-Grid, each surfel can find and connect with its neighbor surfel directly, guaranteeing high efficiency in the loop construction process. Examples are used to illustrate the higher accuracy and better efficiency obtained using this approach.