Architected cellular materials are a class of materials with cellular architecture-dependent properties. Designing the cellular architecture paves a way to generate architected cellular materials with specific properties. However, most of previous studies are based on forward design strategy where a geometry is generated by computer-aided design modeling and its properties are investigated by experiments or simulations. This requires experienced designers and extensive trial-and-error efforts to achieve the desired properties. Consequently, the forward design approach hinders practical applications to some extent. For instance, in tissue engineering, bone implants should be chosen to mimic damaged bones in terms of their biocompatibility, relative density, and stiffness. In such situations, the desired approach is the inverse design method, based on which implants can be designed and generated according to target properties and specific requirements. In this work, we propose an inverse design framework for architected cellular materials using deep learning. This inverse design framework is a three-dimensional conditional adversarial generative network (3D-CGAN) that is trained based on supervised learning using a dataset consisting of Voronoi lattices and their corresponding relative densities and Young’s moduli. The well-trained 3D-CGAN admits variational sampling to generate multiple distinct architectures with target relative density and Young’s modulus. The mechanical properties of 3D-CGAN generated architectures are validated by uniaxial compression tests and finite element simulations. The inverse design framework has a potential application in bone implants where scaffold implants can be automatically generated with target porosity and stiffness.