Abstract The seismic response of gravity dams is typically derived under a deterministic finite element model for the dam-reservoir-foundation system. In the case where uncertainty in material properties should be incorporated into overall dam performance, the sensitive parameters can be treated as random variables. This paper presents the results of a study that considers the spatial distribution of random variables in the context of random field theory. Koyna Gravity Dam is used as a setting for numerical simulations. The concrete modulus of elasticity, mass density and tensile strength are all assumed to be random fields and generated based on the covariance matrix decomposition and midpoint discretization techniques. The anatomy of the random field seismic responses are presented first, followed by a set of parametric analyses. The impact of correlation length, a single- vs. double-random field, one- or two-dimensional material distributions, ground motion intensity and record-to-record variability and, lastly, dam class are all investigated herein. The uncertainty and dispersion of the seismic responses are quantified in each model; it is found that concrete heterogeneity affects the seismic performance evaluation and should be considered in a structural assessment and risk analysis.