Isotopic variation within food sources adds uncertainty to models intended to reconstruct trophic pathways. Understanding this variation is pivotal for planning sampling protocols for food-web research. This study investigates natural variation in C and N stable isotopes among plant species in two western Amazon flooded forests with contrasting watershed biogeochemistry (white-water vrzea-forest and black-water igap-forest). Our objectives were to compare 13C and 15N of leaves and fruits between sites; assess the magnitude of within-site variation in 13C and 15N of leaves (vrzea: 28 spp., igap: 10 spp.) and fruits (vrzea: 22 spp., igap: 22 spp.); determine within-plant variation in 13C and 15N of leaf, wood and fruit tissues; and test whether inter-specific variation in 13C and 15N influence the results of a mixing model predicting the contribution of terrestrial C sources to an aquatic consumer. Mean 13C values of leaves and fruits were not statistically different between the two sites despite regional differences in biogeochemistry and floristic composition. In contrast, mean 15N of leaves and fruits were significantly lower at the vrzea than at the igap site. The high floristic diversity of both forests was reflected in large within-site interspecific variation in both 13C and 15N. Paired comparisons revealed that 13C of wood and fruits and 15N of fruits were generally greater than values obtained for leaves from the same plant. The predicted contribution of different carbon sources to the consumer biomass changed between models as a function of source variability. We discuss implications of source variation for designing sampling protocols, interpreting isotopic signatures, and establishing trophic links between plants and consumers. Our findings highlight the importance of in situ sampling to establish reliable primary production baselines for local food webs.