Researching on the mathematic model of surface roughness in machining hardened steel was to provide the reference for the surface roughness prediction. The contrast experiments of dry turning hardened steel were carried out with ceramic tool (CC6050), cubic boron nitride tool (CB7025) and kentanium tool (GC2025), the surface roughness was measured using 2025 surface roughness tester, the predicted models of the surface roughness were built by Particle Swarm Optimization (PSO) algorithm, the reliability analysis was given out, the shape of chips was observed by scanning electron microscope (SEM). Results proved: the reliability of the predicted models built by PSO was to be verified, it could reflect the relation between the surface roughness and cutting parameters exactly. The feed rate was found out to be dominant factor on the surface roughness in turning with three tools. The saw-tooth chips could decrease the cutting temperature and improve the surface quality.