Rarefaction Curves are frequently used in Environmental Impact Assessments to evaluate sampling sufficiency, but without clear guidelines of how to ensure that the assumptions of the methods are met. Infrastructure projects in the Brazilian Amazon and elsewhere often occupy extensive areas in remote locations with difficult access, and random sampling under such conditions is impractical. We tested the influence of sampling unit (sample or individual), and geographic distance between samples on rarefaction curve s, and evaluated the magnitude of errors resulting from the misuse of rarefaction curve in decision making, using frogs from four Amazonian sampling sites. Individual-based rarefaction curve were steeper than those generated by sample-based rarefaction curve. Geographic distance influenced the number of exclusive species in a predictable fashion only in one area, and not in the Environmental Impact Assessment site. Misuse of rarefaction curve generated large errors in the identification of vulnerable taxa. Because the rarefaction curve model is sensitive to the assumption of randomness and geographic distance can influence it unpredictably, we suggest that rarefaction curve should generally not be used to estimate sample completeness when making management decisions for environmental licensing purposes.