Dengue incidence occurs predominantly within city limits. Identifying spatial distribution of the disease at the local level helps formulate strategies to control and prevent the disease. Spatial analysis of counting data for small areas commonly violates the assumptions of traditional Poisson models due to the excessive amount of zeros. This study compared the performance of four counting models used in mapping diseases: Poisson, negative binomial, zero-inflated Poisson, and zero-inflated negative binomial. The methods were compared in a simulation study. The models analyzed in the simulation were applied to a spatial ecological study of dengue data aggregated by census tracts in the city of Campinas, São Paulo State, Brazil, 2007. Spatial analysis was conducted with Bayesian hierarchical models. The zero-inflated Poisson model showed the best performance for estimating relative risk of dengue incidence in the census tracts. © 2016, Fundacao Oswaldo Cruz. All rights reserved.328