Li, Z.X., Ma, J.H., Qi, Z., Zhang, L.W., Xie, F. and Yang, J.L., 2023. Accelerating the U-net based adaptive subtraction with transfer learning for removing seismic multiples in field data. Journal of Seismic Exploration, 32: 373-384. Adaptive subtraction is essential for removing multiples effectively after the modeling of seismic multiples. The U-net based method has been proposed to conduct adaptive subtraction under the frame of non-linear regression. Compared with the linear regression method the U-net based method can remove more residual multiples and better protect primaries at the cost of higher computational cost. To accelerate the U-net based method for processing the field data efficiently we introduce transfer learning into the U-net based method in this paper. The adjacent seismic gathers have similar mismatches between the modeled multiples and true multiples. On the basis of the transfer learning theory the network parameters of U-net estimated in the previous data part can be used as the initial network parameters of U-net for the next data part. In this way the U-net based method can be accelerated with decreased epoch numbers. While achieving similar accuracy the accelerated U-net based method can decrease the computational time by 40% compared with the non-accelerated U-net based method without transfer learning in the field data example.