In Chinese calligraphy cao shu is regarded as a kind of free form art which differs from other styles greatly in its less constrained strokes and brush textures. In this paper we present a framework for synthesizing cao shu realistically. In our system, we adopt different brush texture patches (BTP) collected from hand-written artworks to represent the solid and hollow strokes appearing in cao shu. BTP can be extended to fit the required length by use of texture synthesis, smooth transitions between adjacent BTP over the stroke are achieved by a Markov Random Field (MRF)-based interpolation technique. With a few parameters we can reproduce typical stroke forms of cao shu, variations of brush texture, the ink amount as well as wetness of the stroke. Our model is also able to synthesize characters of cao shu with different levels of detail that are applicable to scalable font design, very high-quality publishing, computer-aided education and electronic practice of calligraphy.