A semi-empirical prediction method for trailing edge noise is applied to calculate the noise from two modern large wind turbines. The prediction code only needs the blade geometry and the turbine operating conditions as input. Using detailed acoustic array and directivity measurements, a thorough validation of the predictions is carried out. The predicted noise source distribution in the rotor plane (as a function of frequency and observer position) shows the same characteristics as in the experiments: due to trailing edge noise directivity and convective amplification, practically all noise (emitted to the ground) is produced during the downward movement of the blades, causing an amplitude modulation of broadband aerodynamic blade noise at the blade passing frequency (|Dsswish|DS). Good agreement is also found between the measured and predicted spectra, in terms of levels and spectral shape. For both turbines, the deviation between predicted and measured overall sound levels (as a function of rotor power) is less than 1-2 dB, which is smaller than the scatter in the experimental data. Using a smoothed analytical trailing edge noise directivity function, the turbine noise directivity is predicted within 1-2 dB, and the swish amplitude in different directions within 1 dB. This semi-empirical directivity function shows similar characteristics as the theoretical directivity function for a flat plate, except for regions close to the plane of the blade. The validated prediction code is then applied to calculate noise footprints of the wind turbine as a function of rotor azimuth. These footprints show that for cross-wind directions the average level is lower than in the up- and downwind directions, but the variation in level is larger. Even at large distance, swish amplitudes up to 5 dB can be expected for cross-wind directions.