This paper presents an adaptive algorithm for digital watermarking of still colour images. The watermark is embedded in the DCT coefficients of the green channel of the RGB colour image. The algorithm takes into account the perceptual capacity of each coefficient inside the DCT blocks before embedding the watermark information. Therefore, the first 16 low frequency coefficients (excluding the DC value) in the 8×8 DCT block was screened and the eight coefficients with the maximum magnitudes were selected for embedding. The algorithm used is blind and does not require the original image for extracting the watermark. The peak signal to noise ratios (PSNR) is around 60dB. The watermarking method is shown to be robust against JPEG compression, additive noise, cropping, scaling, low-pass and median filtering. © 2009 IEEE.