An intelligent sensor system for on-line estimation of surface roughness in the grinding process is developed. The system consists of a statistical signal processing algorithm and a neuro-fuzzy model of surface roughness. The model is established using a set of experimental data. The sensor signals of acoustic emission CAE), normal force and tangential force generated in grinding zone are acquired online in real time. The surface roughness of the ground workpiece is measured offline and used as the target output of the model. A first order Sugeno-type neuro-fuzzy inference system is employed to optimize the model by minimizing a sum of rootmean- squared residuals. The results show that the system can achieve a satisfactory perrformance for on-line estimation of surface roughness.