With 15% sampling ratio, two ways in dividing groups, three ways in determining sampling numbers in each group, and two methods for selecting individuals, 18 agronomic characteristics of 476 sweet potato collection in Guangxi province were analyzed and 13 samples of primary core collection were developed. To determine the effectiveness, each of primary core collection was compared with the whole collection for 5 parameters including ratio of phenotype retained, variance of phenotypic frequency, index of genetic diversity, coefficient of variation and coincidence rate of range. The result showed that the way of according to the resources type was significantly better than according to the source in developing core collection. Among the methods to determine the sampling numbers in each group, logarithm method was the best, followed by proportioned and square root method. On the selection of individuals, least distance stepwise sampling method was better than random method. Therefore, the optimum sampling strategy for developing sweet potato core collection is firstly to group by resource type, then determine the sampling numbers in each group by proportion of logarithm, and finally choose individuals through least distance stepwise sampling method.