The invention discloses a real-time image automatic white balance system and method based on an FPGA, and the method comprises the steps: carrying out the statistics of R, G and B information; judging whether the color of an image is abundant or not; selecting a corresponding threshold value fv for color temperature estimation; selecting pixels with R, G and B being all greater than fv as reference points; employing the mean value of R, G and B of the reference pixels for gain calculation; and determining the adjustment coefficients u, v and w of the R, G and B. The method carries out the statistics of mean values of R, G and B of one image two times, carries out one-time buffering of the image in order to reduce time delay, calculates the threshold value fv during buffering, reads the buffered image for interpolation, calculates the adjustment coefficients u, v and w of the R, G and B, and adjusts the next image. The method integrates the advantages of a gray world algorithm and a total reflection method, is small in hardware cost, is high in adaptability, and is good in white balance effect.