A new algorithm is presented to extract colored texture by effectively merging the texture feature, color feature and spatial correlation of color texture based on wavelet decomposition. Experiments are conducted on a set of 20 natural colored texture images in which multiple feature fusion and classification can be performed on the basis of the pyramid wavelet decomposition (PWD), incomplete tree-structured wavelet decomposition (ICTSWD) and wavelet packet decomposition (WPD). It is demonstrated that correct class rate of multiple feature fusion based on PWD is 85.78% and correct class rate based on WPD is 91.03% with the dimensionality increased exponentially, but the dimensionality of feature fusion based on ICTSWD descended greatly because of selective decomposition in sub-band, and correct class rate is 90.63% after fusion, simultaneously, multiple feature fusion based on ICTSWD has better anti-noise ability than fusion using WPD.