In this study, using in-process tool failure detecting system by AE method in turning machining, we measured AE signal from the tool, and the surface roughness of workpiece and then compared it with tool wear. As a result, we found that tool failure can be predicted by means of surface roughness of the workpiece and it can be predicted more precisely by the arithmetical average roughness (Ra) than by the maximum height of irregularities (Rmax) of the workpiece. Also, we found that we could judge whether it was sudden failure or the wear by means of the shape of AE signal and the range distri- bution of power spectrum frequency when tool danage was happened.