Tropical forests are an important global source of volatile organic compounds (VOCs) and other atmospheric trace gases. The high biodiversity in tropical rainforests complicates the extrapolation of biogenic volatile organic compound (BVOC) emissions from leaf-level measurements to landscape and regional or global scales. In Amazônia, a significant fraction of the carbon emitted from the biosphere to the atmosphere is emitted in the form of BVOCs, and the knowledge of these emissions is important to our understanding of tropical and global atmospheric chemistry and carbon cycling. As part of the Large scale Biosphere-atmosphere experiment in Amazônia (LBA). VOC concentrations were measured at two sites near Santarém, Para State, Brazil. The two sites are located in the National Forest of Tapajós, the first corresponding to primary forest and the second to a forest, that was selectively logged. The samples were collected simultaneously at heights of 65 and 55 m (20 and 10 m above forest canopy, respectively). The average isoprene mixing ratio was 2.2–2.5 ppb at the two sites and the standard deviations within a site ranged from 1 to 1.2 ppb. A strong seasonality of isoprene mixing ratio was observed and associated with the wet and dry seasons. The lowest mixing ratios were found during the transition between the wet to dry season, while at the start of the biomass burning season the mixing ratios increase. A qualitative analysis of a one dimensional model demonstrates that the seasonal cycle in concentrations reflects changes in isoprene production by the ecosystem, not changes in boundary layer dynamics or chemistry. The magnitude of the cycle indicates that the physiological capacity of the ecosystem to emit isoprene may depend on water availability although phenological changes could also contribute to the observed variations. A simple 1-D model that assumes mean daytime isoprene fluxes of 1.5 mg m^−2h^−1 and 0.9 mg m^−2h^−1 scaled by an algorithm depending on precipitation at the primary forest and selectively logged sites, respectively, is able to reproduce the observed vertical gradients.