Three in-process surface roughness prediction (ISRP) systems using linear multiple regression, fuzzy logic, and fuzzy nets algorisms, respectively, were developed to allow the prediction of real time surface roughness of a work piece on a turning operation. The surface roughness is predicted from feed rate, spindle speed, depth of cut, and machining vibration that is detected and collected by an accelerometer.;Two groups of data were collected for two cutters with nose radii of 0.016 and 0.031 inches, respective. A total of 162 training data sets and 54 testing data sets for each cutter were applied to train and test the system. While the multiple-regression-based system applied the linear relationships of the dependent variables and the dependent variable for the prediction, the fuzzy-logic-based and the fuzzy-nets-based systems relied on fuzzy theory for the prediction. The fuzzy rule banks employed in the fuzzy-logic-based system was generated with expert's experiences as well as observation results from the experiments. Whereas, the rule banks employed in the fuzz-nets-system were rule banks self-extracted from the training data by the fuzzy-nets self-learning algorithm.;The predicted surface roughness values were compared with corresponding measured values. The average prediction accuracy with the three algorithms, linear multiple regression, fuzzy logic, and fuzzy nets algorisms, was 92.78%, 89.06%, and 95.70%, respectively. The use of the accelerometer was found valuable in increasing the prediction The Fuzzy-nets-based In-process Surface Roughness Prediction System was considered the best among the three tested systems. This conclusion relies on not only the best average prediction accuracy achieved, but also the self-learning ability of the fuzzy nets algorism.