Spatially detailed epidemic models commonly invoke probabilistic cellular automata to predict population-level consequences of localized interactions between infectious and susceptible individuals.  Most such models equate local and global host density; the resulting spatial uniformity implies that each individual interacts with the same number of neighbors.  However, many natural populations exhibit a heterogeneous spatial dispersion, so that the number of contacts capable of transmitting an infection will vary among interaction neighborhoods.  We analyze the impact of this variation with a probabilistic cellular automaton that simulates a spatial epidemic with recovery.  We find that increasing spatial heterogeneity in host density decreases the frequency of infection at endemic equilibrium, and consequently increases the divergence between mean-field predictions and observed levels of infection.