In this paper, first, the relation between roughness and grinding parameters is built. Then, an intelligent dynamic identification model of surface roughness is developed, which bases on the theory of roughness during grinding and the theory of fuzzy-neural network. The inputs for the model are the grinding parameters. Besides, an accelerometer is used to gather the dynamic vibration signal in real time. The model was used in the external cylindrical grinding experiment, and the result verified that the proposed model was feasible.