Abstract: An algorithm was proposed to predict the weld seam deviation in high-power fiber laser(maximal laser power 10kW) welding of type 304 austenitic stainless steel.A highspeed camera was employed to capture the infrared-images of the molten pool in welding process.The eigenvectors such as keyhole centroid,keyhole configuration parameter,heat accumulation effect parameter and so on reflected the deviations between the laser beam and the weld seam position,which were applied as the inputs of a RBF(radial basis function) neural network,and a RBF neural network model was established to predict the weld deviations.The eigenvectors of weld deviations were sampled to train the prediction model,and the established prediction model was tested by the fiber laser welding data.Experimental results showed that the founded model could predict the deviations between the laser beam and the weld seam position during the highpower fiber laser welding.