Abstract: To study the genetic variation of main agronomic traits of millet, to reveal the main factors affecting grain weight per spike of millet, and to screen the lines with good comprehensive traits can provide theoretical basis and recommended materials for breeding of high-yield hybrid millet varieties.In this study, 341 recombinant inbred lines（RILs） constructed by selfing separation of Zhangzagu 3（sterile line A 2×Jizhanggu 1×SR 3522） were used as materials to conduct genetic variation analysis, correlation analysis and stepwise regression analysis, path analysis and cluster analysis based on 16 agronomic traits including heading period, penultimate leaf length, penultimate leaf width, plant height, stem diameter, number of elongation node, stem number of spike, neck-panicle length, ear length, ear diameter, ear yardage, code grain number, ear weight, kernel weight of spike, grass weight per plant and 1 000-grain weight.The results showed as follows 1) there were extremely significant differences in stem number of spike, ear weight and kernel weight of spike of parents.In addition, the variation of main agronomic traits in RILs population was abundant, and the distribution frequency was approximately normal, indicating bidirectionally transgressive segregation. 2) correlation analysis showed that the kernel weight of spike was positively correlated with heading stage, stem number of spike, grass weight per plant and ear weight, and negatively correlated with penultimate leaf length, stem diameter, number of elongation node and 1 000-grain weight. 3) the fitting degree based on multiple regression analysis was R~2=0.920,indicating that the traits investigated in this study basically covered the main traits affecting the kernel weight of spike.Meanwhile, path analysis showed that 1 000-grain weight, ear weight, stem number of spike, ear diameter, ear length, heading stage, penultimate leaf length, penultimate leaf width, neck-panicle length, number of elongation node, code grain number, stem diameter, ear yardage and plant height were the main factors affecting the kernel weight of spike per plant. Among them, heading stage, stem number of spike, ear length, ear diameter, ear weight and 1 000-grain weight had significantly positive and direct effects on the kernel weight of spike, while penultimate leaf length, penultimate leaf width, plant height, stem diameter, number of elongation node, neck-panicle length, ear yardage and code grain number had significantly negative and indirect effects on the kernel weight of spike per plant. 4) at the Euclidean distance of 25.0, the RILs population could be divided into 6 groups. The group Ⅰ consisted of 269 materials, which had a longer growth period and lower yield, while the group Ⅱ consisted of 60 materials with low yield. Group Ⅲ consisted of 8 materials and had the lowest yield. Group Ⅳ consisted of two materials, and the yield of this group was moderate. Group Ⅴ included 1 material, and the yield of this group was superior level.Additonally, the yield of group Ⅵ was middle and upper level, including 1 material.The results provided a reference for selecting excellent planting resources and selecting hybrid parents.