Many biological systems approach physical limits to their performance, motivating the idea that their behavior and underlying mechanisms could be determined by such optimality. Nevertheless, optimization as a predictive principle has only been applied in very simplified setups. Here, in contrast, we explore a mechanisticallydetailed class of models for the gap gene network of the Drosophila embryo, and determine its 50+ parameters by optimizing the information that gene expression levels convey about nuclear positions, subject to physical constraints on the number of available molecules. Optimal networks recapitulate the architecture and spatial gene expression profiles of the real organism. Our framework makes precise the many tradeoffs involved in maximizing functional performance, and allows us to explore alternative networks to address the questions of necessity vs contingency. Multiple solutions to the optimization problem may be realized in closely related organisms.