By the very nature of its origins and deposition, geotechnical materials such as soils and rocks are spatially variable in their engineering properties. This variability is rarely taken into account directly in traditional geotechnical analysis, rather some "representative average" or "appropriately conservative" property is assumed to act across the whole region of interest. This problem has drawn considerable attentions in geotechnical engineering community and the problem of modelling of spatial variability is becoming an active research area. Recent studies showed that ignoring spatial variability of soil/rock may lead to unsafe designs. This significant finding has been proved by comparing reliability results of traditional probabilistic methods and state-of-art random finite element method in 1 -, 2- and 3-dimensions. Most recently, a new framework of quantitative risk assessment of landslides is proposed, which is based on the logic that consequence should be assessed individually for each failure mode. This framework is generally applicable to other types of risk assessment in civil engineering.