In this paper, narrow-band interference (NBI) mitigation is addressed in amplify-and-forward orthogonal-frequency-division-multiplexing (AF-OFDM) cooperative communication systems. Based on the channel gains between the interferer, destination and the relay nodes, three copies of the NBI are received at the destination node in addition to the desired signal. Hence, NBI degrades the performance of AF-OFDM systems which motivates the need for mitigation techniques to reduce its effect. NBI is a sparse signal in the frequency-domain, hence, compressive sensing (CS) framework can be used to estimate and cancel the NBI before detecting the transmitted signal. However, frequency-grid-mismatch destroys the sparsity of NBI in the frequency domain at the destination terminal. A structured-dictionary-mismatch formulation is proposed to approximate the received NBI vector by two sparse vectors. An £2,1 norm minimization problem is solved to recover the sparse vectors. The recovered NBI is then canceled from the received signal before detection. Simulation results demonstrate the merits of the proposed approach.