This paper investigates the probabilistic nature of ultimate limit state failures of deep foundations in purely frictional soils (e.g., sands). In so doing, the theory required to predict both the probability of ultimate limit state failure and the resistance factors needed to avoid this limit state are proposed. The proposed resistance factors are functions of site understanding and failure consequence, and the theory leading to these resistance factors is validated via Monte Carlo simulation of a two-dimensional spatially variable random field. In both the theory and the simulation, a pile is assumed to be placed vertically at a certain position in the soil mass, and the soil is sampled at various distances from the pile to come up with characteristic soil properties (namely friction angle) for use in the pile design. Agreement between theory and simulation is found to be very good. The theoretical model is then employed to determine upper bound geotechnical resistance factors, which can be used to complement current ultimate limit state design code calibration efforts. An example of such a calibration is presented.