A sesame core collection containing 453 accessions was established by assessing genetic diversity of 4251 accessions collected in China. The accessions in the whole collection were grouped based on their origin, varietal and agro-ecological type. A computer program was used to cluster the data of 14 traits from the random small sample. Ward's method was chosen as the desirable method for clustering sesame germplasm resources. A set of technical methods for establishing core collection has been developed through determining the number of selected accessions based on the diversity distribution. The agronomic traits of the pre-selected core collection were investigated and tested in three ecological locations in China for two years. Among six qualitative traits, including plant type and flower color in a total of 5304 accessions, 97.85% was completely the same as the original data, and three quantitative traits, including growth period and plant height, were significantly and positively related with the original data. The representativeness assessment of the final core collection indicated that 35 phenotypes of 10 qualitative traits in the core collection were basically similar to those of the pre-selected core collection. Six feature values such as mean, standard deviation etc. of four quantitative traits including oil content and 1000-seed weight were close to those of the pre-selected core collection. The final core collection possessed a fair representativeness of the pre-selected core collection.