The invention relates to a method for predicting far-field noises of a converter transformer. The method comprises the steps of 1) according to a factory sound power level and a noise spectrum of theconverter transformer, building a Cadna/A prediction software-based model; 2) measuring a near-field noise value of the converter transformer and a spectrum of a sound receiving point; 3) converting sound pressure levels into sound pressures, and calculating respective contribution value proportions eta, wherein the sound pressure levels include sound pressure levels LPi, contributed by a converter transformer body and a fan of the converter transformer body, of the sound receiving point and a total sound pressure level LP of the sound receiving point; 4) by applying an inversion method of Cadna/A software, according to a contribution value proportion, adjusting sound power levels of the converter transformer body and the fan to make a predicted value of the sound receiving point basicallyconsistent with an error of an actually measured value; and 5) according to the adjusted model, performing far-field noise prediction of the converter transformer. According to the method, through Cadna/A software modeling, the model is adjusted to be matched with noise distribution around the converter transformer according to the percentage of the contribution values of the noise spectrums of the converter transformer body and the fan, thereby realizing accurate prediction of the far-field noises of the converter transformer.