This study attempted to build up a core collection of 1285 sorghum germplasm accessions at Shanxi Gene Bank. These accessions were historically investigated for the phenotypic variations 18 agronomic traits， and these datasets were subjected for tests using the combinations on sampling methods， sampling proportion and clustering methods. A pipeline combining multiple cluster deviation sampling method， 15% sampling proportion， euclidean distance and complete method was used. By pairwise comparison of phenotypic datasets in core collection（ 192 accessions） to all samples， the mean difference percentage（ MD） was 0， the variance difference percentage（ VD） was 83.33%， the coincidence rate of range（ CR） was 97.45%， and the variable rate of coefficient of variation（ VR） was 119.63%. Furthermore， eight accessions with particular characters were incorporated. This core collection including 198 landraces has been proposed， followed by tests using t-test， extreme value， standard deviation， the t-test of Shannon-Weaver diversity index， as well as principal component. As a result， 198 sorghum landraces are representative on the diversity of sorghum landraces hosted by Shanxi Gene Bank， and this collection might be useful in practical uses for sorghum breeding.