Abstract: As a major starch-based raw material for fuel ethanol production, sweet potato Ipomoea batatas (L.) Lam. is a critical industrial material and a new energy resource. The development of new sweet potato varieties with high starch yield has been a key element of sweet potato research. To improve breeding efficiency of sweet potato varieties with high starch yield, short breeding cycle and valuable index supply for early selection of high starch yield sweet potato breed, this study used the Unweighted Pair Group and PSC- (Pair Similarity Coefficient) based Arithmetic Average Method to classify 48 main sweet potato germplasm resources in terms of starch yield, morphological traits and synthetic enzyme activity. Then correlation analysis was used to assess the relations of starch yield with agronomic traits and starch synthetic enzyme activity at different times. The results showed significant variations in agronomic traits of different sweet potato varieties/lines and periods. Cluster results indicated that agronomic traits in 100 days after planting were most relevant with starch yield. Among the agronomic traits, starch yield was significantly negatively correlated with plant branch (r = 0.428) and significantly positively correlated with dry matter rate (r = 0.423) in 100 days after planting. No significant correlation existed between starch yield and maximum vine length, root tuber number per plant or root tuber fresh weight per plant. It was possible to use plant branch and dry matter in 100 days after planting as indexes in early selection of high starch yield sweet potato breed. The cluster results for key starch synthesis enzymes (ADPG-PPase, SS and SPS) activities at different times were significantly different. The cluster results showed key starch synthesis enzymes activities in 50 days after planting was most relevant with starch yield. Among the enzymes activities in 50 days after planting, starch yield was negatively correlated with ADPG-PPase activity (r = 0.163), and positively correlated with SS (r = 0.101) and SPS (r = 0.016) activities; all of which were insignificant. Because of the weak correlation and tedious determination process, it was not possible to use the activities of ADPG-PPase, SS and SPS as physiological indexes in early selection of high starch yield sweet potato breed. Results of the starch yield and agronomic traits analyses provided valuable reference for screening breeding materials in developing new high starch yield varieties. On the basis of existing studies on agronomic traits, it was possible to use high dry matter rate and lower plant branch within 100 days after planting as reference indexes in early generation in breeding sweet potato high starch yield varieties.