Abstract: Pore network controls important engineering properties of the soil such as seepage,drainage consolidation and matrix suction. This article introduces the basic principle and algorithm of the maximal ball algorithm used in establishing the pore network model of the soil space. Taking the Nanjing silty sand specimen fabricated by vibrating-dry method and scanned by micro-CT as an example,we use the maximal ball algorithm to extract the spatial pore network ball-stick model for Representative Element Volume(REV)in the three-dimensional reconstruction model of the specimen,and calculates the pore network parameters at REV scale. Statistical results shows that the pore parameters such as pore radius,throat radius,pore coordination number,area shape factor of pore,area shape factor of throat and throat length approximately obey the normal distribution,the pore volume approximately obeys the attenuation exponential distribution. The pore radius and throat radius distributes within 100 μm and 65 μm respectively,and their mathematical expectations are 40 μm and 18 μm; the pore coordination number distributes within 25 and the mathematical expectation is 5.1; the area shape factor of pore and area shape factor of throat distributes in the range of 0.01~0.04 and 0.01~0.05 respectively,and their mathematical expectations are 0.019 and 0.033; the throat length distributes in the range of 100~800 μm and the mathematical expectation is 292.22 μm. Meanwhile,we find that the number of small pore with volume less than1.5×10⁷ μm³ is more than 90%. This method can be used to the quantitative characterization of micro pore structure of soil.