This paper presents an exercise that estimates a tax function for the recurrent tax on immovable property by stochastic frontier analysis with panel data for the Brazilian municipalities during 2002-2014. It is used two procedures to deal with heterogeneous observations. Firstly, an adaptation for panel data of the approach used for excluding outliers in Sousa, Cribari-Neto e Stosic (2005). A second procedure is to estimate separate regressions for more comparable groups of municipalities, following the typology of urban cities hierarchy. The results suggest an additional revenue of 0.34% of the GDP if most of municipalities would approximate their tax effort to those with the best performance within each of the homogeneous groups. This could be an alternative for strengthening fiscal conditions and for reducing the dependency on grants revenues and on anti-growth taxes.