In random finite element analysis (RFEA), continuous random fields must be discretized. The critical element size to achieve acceptable accuracy in effective Young’s modulus for an elementary soil mass is investigated. It is observed that the discrepancy between the continuous and discretized solutions is governed by the discretization strategy (element-level averaging versus midpoint), spatial variability pattern, and the adopted autocorrelation function. With the element-level averaging strategy, RFEA with element size less than (scale of fluctuation)/5 will not induce significant discrepancy from the continuous solution. Moreover, the element-level averaging strategy is more effective than the midpoint strategy.