Pyramidal cells show marked variation in their morphology, including dendritic structure, which is correlated with physiological diversity; however, it is not known how this variation is related to a cell's role within neural networks. In this report, we describe correlations among electrosensory lateral line lobe (ELL) pyramidal cells' highly variable dendritic morphology and their ability to adaptively cancel redundant inputs via an anti-Hebbian form of synaptic plasticity. A subset of cells, those with the largest apical dendrites, are plastic, but those with the smallest dendrites are not. A model of the network's connectivity predicts that efficient redundancy reduction requires that nonplastic cells provide feedback input to those that are plastic. Anatomical results confirm the model's prediction of optimal network architecture. These results provide a demonstration of different roles for morphological/physiological variants of a single cell type within a neural network performing a well-defined function.