Most rare species appear to be specialists in plant–pollinator networks. This observation could result either from real ecological processes or from sampling artifacts. Several methods have been proposed to overcome these artifacts, but they have the limitation of being based on visitation data, causing interactions involving rare visitor species to remain undersampled. We propose the analysis of food composition in bee trap nests to assess the reliability of network specialization estimates. We compared data from a plant–pollinator network in the Monte Desert of Villavicencio Nature Reserve, Argentina, sampled by visit observation, and data from trap nests sampled at the same time and location. Our study shows that trap nest sampling was good for estimating rare species degree. The rare species in the networks appear to be more specialized than they really are, and the bias in the estimation of the species degree increases with the rareness. The low species degree of these rare species in the visitation networks results from insufficient sampling of the rare interactions, which could have important consequences for network structure.