A total of 3 170 strains of rice germplasm from 20 countries were studied by refer to similar studies native and abroad. A core collection of these germplasm was constructed. The results were as follows: One billion basal data of 25 classifying trains of 3 170 varieties were obtained. The optimum method was determined after 4 years’ experiment. That is sampling 15 percent of total amount, grouping according the code, simple proportional sampling and clustering sampling with group. A core collection bank of 447 varieties out of 3 170 varieties was constructed. These varieties, accounted for 15% of total germplasm, could represent the total germplasm bank. All data could be treated with computer automatically.