Abstract: Uncertainty characteristic of geotechnical parameters summarized from large sample data can provide an objective basis for calculating indexes in reliability analysis of geotechnical engineering and improving the precision of calculation results. Test data from nearly one thousand geotechnical specimens were collected from six railways in China. According to the cluster analysis of engineering geological units, 13 groups of large samples of soil samples were established, and the probability distribution of soil strength parameters(c, φ) and their coefficients of variation (δ_c, δ_φ) were analyzed using probability and statistics method. The ranking of soil physical-mechanical indexes affecting the variation of strength parameters was discussed using Gray Relational method. The results show that: δ_c and δ_φ generally present large variation level, and the value of δ_c and δ_φ are positive correlated. The state indices of soil such as liquidity index I_L, water content ratio w, coefficient of compressibility a_1-2 and natural porosity index e present more relatively influence to δ_c and δ_φ than the plasticity index I_P, which reflects the soil property. The index of I_L is the most important correlation factor affecting δ_c and δ_φ, and then followed by a_1-2 and w respectively. The correlation between variation level and impact factors can provide prior information for judgement of geotechnical uncertainty in a limited sample case.