When a huge number of couplers are required in a large-scale beam forming network (BFN), an efficient coupler automatic design method is in demand. In this letter, a new data-driven-based modular neural network (DD-MNN) is proposed to address this problem. The framework consists of five submodules containing artificial neural networks (ANNs). First, training samples are divided into satisfying and unsatisfying couplers. Second, neural networks in submodules are trained to learn the deep relations between electrical and geometrical parameters from satisfying couplers. Third, the learned knowledge from satisfying samples is applied to unsatisfying ones to provide optimization suggestions. Finally, data are returned back to drive the fine-tuning of ANNs. An experiment using 75 four-port couplers is designed, showing that the method can output the optimized structure with over 95% improvement. Finally, the framework is implemented to design ten couplers given <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$S_{31}/S_{41}$ </tex-math></inline-formula> , showing that the proposed method could perform an efficient coupler automatic design.