Virtual neurons are essential in computational neuroscience to study the relation between neuronal form and function. One way of obtaining virtual neurons is by algorithmic generation from scratch. However, a main disadvantage of current available generation methods is that they impose a priori limitations on the outcomes of the algorithms. We present a new tool, EvOL-NEURON, that overcomes this problem by putting a posteriori constraints on generated virtual neurons. We present a proof of principle and show that our method is particularly suited to investigate the neuronal form-function relation. (c) 2007 Elsevier B.V. All rights reserved.