No standard techniques yet exist for assessing the predictive performance of models that simulate spatially-explicit processes. Spatial simulation models generate autocorrelation patterns that are a critical aspect of model prediction. These patterns can be quantified using spatial (Moran's I) and spatio-temporal (Griffith's STI) autocorrelation indices. A method of error analysis for spatial simulation models is demonstrated using autocorrelation indices to distinguish among the effects of parameter uncertainty on a stochastic spatial simulation of seed dispersal. The resulting autocorrelation patterns of state variables display a range of nonlinear, counterintuitive effects. In contrast to techniques proposed for spatial model error analysis (contagion, spatial predictability, adjacency, multiple resolution procedure, fractal dimension, interface), autocorrelation indices can be used with interval-scaled data and have well-defined sampling distributions that enable significance testing.