This paper presents a performance-based reliability design methodology for axially loaded drilled shaft foundations using the Monte Carlo statistical methods. The performance criteria for an axially loaded drilled shaft are defined in terms of the drilled shaft head displacement. The load transfer method is used as the computational model for solving the nonlinear soil–foundation interaction problem and to predict the load–settlement curve at the drilled shaft head. The input to the computational model and the model error are treated as random variables. Particularly, soil properties such as soil strength parameters are modeled as random fields to account for soil spatial variability. Random field samples are generated using local averaging subdivision according to the prescribed statistical descriptors, including mean, variance, and correlation length. Given performance-based acceptance criteria, the probability of failure can be evaluated for the prescribed load effects. A computer program is developed to facilitate the computation of the load–displacement curves of the drilled shaft head by using the commonly adopted load transfer concepts (t–z curves and q–w curves for shaft side and toe, respectively). Two design examples are presented, one for uplift and the other for compression, to illustrate the application of the developed performance-based reliability design methodology. One of the important observations from the examples is that the computed probability of failure can be sensitive to the spatial variations of soil strengths characterized by the correlation structures. It is strongly recommended that spatial variation of soil properties be considered in the reliability-based foundation design. Copyright © 2012 John Wiley & Sons, Ltd.