Because major capacity-expansion projects are very unlikely in the coming years, transportation planners need to view the existing infrastructure as fixed and to start thinking about how much development the current system can sustain. This line of thinking, which involves deriving land use limits from infrastructure capacity, requires solving the inverse of the typical transportation planning problem. Modular artificial neural networks (ANNs) were developed for solving the inverse transportation planning problem. ANNs were designed to predict zonal trip ends, given the traffic volumes on the links of the transportation network. Computational experiments were performed to study the effect on ANN accuracy of three factors: transportation network size, variability in training data, and ANN topology. ANNs were shown to be quite capable of capturing the relationship between link volumes and zonal trip ends for both small and medium-sized transportation networks and for degrees of variability in the training data. Modular ANNs with one or two hidden layers appeared to outperform other ANN topologies.