To predict the macroscopic properties (e.g., transport, electromagnetic, and mechanical properties) of porous media, it is necessary to have a three-dimensional (3D) representation of porous media. We reconstruct the geologically realistic 3D structure of Fontainebleau sandstone based on the two-dimensional (2D) thin sections by using the multiple-point statistics method. For this method, the size of template is an important parameter that reflects the perceived scale of spatial structure of porous media. In this paper, we take advantage of entropy method to obtain the appropriate size of the template, which is proven to be correct and feasible. The reconstruction method proposed by us combines successive 2D MPS simulations as well as 3D MPS simulation, which takes account into the pore structure information (e.g., heterogeneity and connectivity) both intralayer and interlayer. This reconstruction method is tested on Fontainebleau sandstone for which 3D images from micro-CT scanning are available. Applying local percolation theory analysis, this new approach can depict the expected patterns of geological heterogeneities. In addition, it also can well reproduce a high degree of connectivity of the pore space better than other reconstruction methods based on lower-order statistics. Copyright © 2011 John Wiley & Sons, Ltd.