The magnitude distribution law of DWT coefficients of color image data in Karhunen-Loeve space is investigated. Considering the KLT matrix formed, the magnitude of DWT coefficients in components K1, K2 and K3, in general, have the relation K1 > K2 > K3 where K1, K2, K3 are the three components of a color image in KL space. Based on this characteristic and other DWT coefficient features, a novel ordered-SPIHT for embedded color image coding (OCSPIHT) is proposed in this paper. The set of DWT coefficients of an image in KL space are split into several sets according to their magnitudes. For each specific threshold, only the sets with coefficients equal to or larger than the threshold will be encoded. This can save many bits and improve coding performance especially at low bit rate. © 2003 IEEE.