Five hundred and ninety-six Saccharum. spontaneum L. were selected to investigate the best sampling strategy for pre-core collection. Based on 21 quantitative and 6 qualitative traits, 50 sampling procedures were established concerning grouping principles，sampling proportions within group，selection methods within group. First grouping or non-grouping was determined based on collecting region, altitude, stalk diameter, latitude, zoology region and cluster. Then sampling within group was set up based on simple proportion, logarithmic proportion, square root proportion and genetic proportion. And random and cluster selection was applied for sampling within group. At the same time, 10 grads for sampling (5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, and 50%) from total resource were analyzed to decide the best sampling grad. The sampling strategies were evaluated according to comprehensive score based on indices of genetic diversity, variance of phenotypic value, variance of phenotypic frequency, coefficient of variation and ratio of phenotypic traits retained. The results showed that cluster sampling was better than the random sampling, collecting regions grouping and stalk diameter grouping were the best grouping and genetic proportion was the best sampling proportions within group. Finally, after analyzing 50 sampling strategies and 10 sampling grads, sampling based on stalk diameter grouping, genetic proportion and cluster sampling combined with sampling grad of 15% was confirmed to be the optimal sampling strategy for pre-core collection. A pre-core collection contained 92 samples including 5 additional ones with the rare phenotypic traits was established according to the best sampling strategy, which accounted for 15.44% of total resources. The genetic diversity, variance coefficient, variance of phenotypic frequency and variance of phenotypic value of this pre-core collection were better than that of the total resource and ratio of phenotypic traits retained came to 100%, which suggests the pre-core collection could preferably represent the total resource.