Airfoil self-noise is one of the dominant sources of airframe noises which causes limitations in many applications such as wind turbines. This paper investigates and quantifies the sensitivity of airfoil self-noise prediction to the grid resolution using large eddy simulation. Three-dimensional incompressible fluid flow around a NACA0012 airfoil at zero angle of attack with a chord-based Reynolds number of 6.4×105 is numerically analyzed in this paper. Far-field noise is predicted by Ffowcs-Williams & Hawkings model using the results of large eddy simulation. Three different grid sizes are used to investigate the effect of grid resolution on the accuracy of self-noise prediction. Results are compared with the experimental data of wind tunnel tests and noise measurements with microphones. Although the aerodynamic properties are calculated accurately in all grids, the grid resolution over the surface has a significant effect on the accuracy of the noise prediction. This effect of grid resolution is quantified in this paper. By the increase of grid points in the spanwise and streamwise directions on the surface, numerical noise prediction has approached the experimental data. The difference with the experimental data decreases from 20 dB to 3 dB in some frequencies. In addition, having doubled the number of surface grid points in both directions the average percentage of difference with the experimental data decreases from 5% to 2%.