Power system studies require the topological structures of real-world powernetworks; however, such data is confidential due to important securityconcerns. Thus, power grid synthesis (PGS), i.e., creating realistic powergrids that imitate actual power networks, has gained significant attention. Inthis letter, we cast PGS into a graph distribution learning (GDL) problem wherethe probability distribution functions (PDFs) of the nodes (buses) and edges(lines) are captured. A novel deep GDL (DeepGDL) model is proposed to learn thetopological patterns of buses/lines with their physical features (e.g., powerinjection and line impedance). Having a deep nonlinear recurrent structure,DeepGDL understands complex nonlinear topological properties and captures thegraph PDF. Sampling from the obtained PDF, we are able to create a large set ofrealistic networks that all resemble the original power grid. Simulationresults show the significant accuracy of our created synthetic power grids interms of various topological metrics and power flow measurements.