High uncertainty is an inherent behavior of geotechnical materials. Nowadays, random field theory is an advanced method to quantify the effect of high uncertainty on geotechnical engineering. This study investigates the effect of spatial variable soil layers on deformations of deep excavation via the random finite element method. A procedure based on PLAXIS 2D software was developed to generate two-dimension random finite element models including multiple variables. Via the K-S test and S-W test, the excavation deformations basically followed lognormal distribution. With the growth of standard deviation of soil properties parameters, the distribution of excavation deformations becomes wider, and the failure probability increases. When the vertical scale of fluctuation ranges from 1 m to 25 m, the distribution of excavation deformations becomes wider. To analyze system reliability, this study proposed a fitted multiple lognormal distribution methods, which was a method with higher efficiency. The results indicated that system reliability was lower than single failure probability and sensitive to design level. The system failure probability will be over-evaluated or under-evaluated if the correlation between excavation responses is ignored. This study provided a novel method to quantify the effect of high uncertainty of soil layer on excavation responses and proposed an efficient method for system reliability analysis, which is meaningful for excavation reliability design.