The random field (RF) theory is widely used for describing the soil spatial variability in geotechnical engineering. This article presents soil variability modellings by using different types of RF-based on the available measurements of an earth dam. The effects of these RFs on dam reliability are investigated as well. The studied dam is well-documented, and there are many geo-localized measurements for the dry density. These measurements are firstly used to estimate the basic parameters of unconditional-stationary RFs and are then explored to define two more complex RFs (one is conditional RF and the other considers the mean variation with depth). The three mentioned types of RFs are all implemented in the same reliability analysis procedure for comparison. The results demonstrate that using different RFs for soil spatial modelling would induce insignificant differences in terms of reliability results if the dam construction was well controlled (careful selection for the construction material and controlled compaction). Therefore, a simple RF (unconditional-stationary) is enough to obtain satisfactory results in the case of carefully controlled dams during their construction. Otherwise, conditional RFs are recommended if more accurate results are needed, given that this type of RF is conditioned on the available data and can consider the non-stationarity of a soil property.