This paper aims to generate materials for 3D meshes from text descriptions.Unlike existing methods that synthesize texture maps, we propose to generatesegment-wise procedural material graphs as the appearance representation, whichsupports high-quality rendering and provides substantial flexibility inediting. Instead of relying on extensive paired data, i.e., 3D meshes withmaterial graphs and corresponding text descriptions, to train a material graphgenerative model, we propose to leverage the pre-trained 2D diffusion model asa bridge to connect the text and material graphs. Specifically, our approachdecomposes a shape into a set of segments and designs a segment-controlleddiffusion model to synthesize 2D images that are aligned with mesh parts. Basedon generated images, we initialize parameters of material graphs and fine-tunethem through the differentiable rendering module to produce materials inaccordance with the textual description. Extensive experiments demonstrate thesuperior performance of our framework in photorealism, resolution, andeditability over existing methods. Project page: https://zju3dv.github.io/MaPa