Abstract: This paper proposes to evaluate the local wall thinning of carbon steel pipe from multi-frequency electromagnetic field. Firstly, the feature signals are determined by correlation analysis of the signals and the wall thinning sizes. Subsequently, the models for estimating the residual wall thickness rt and wall thinning length w are constructed respectively using Gaussian process regression (GPR) and linear regression. Finally, the applicability of the models to the evaluation of local wall thinning is verified by simulation and experiment.