Color error diffusion can be classified into two types, namely, vector error diffusion and scalar error diffusion, according to the underlying quantization methods. Compared to scalar error diffusion, vector error diffusion is superior in image quality. However, it requires significantly more computation, and can introduce artifacts due to accumulation of the errors in output device space. In this paper, we propose a new quantization algorithm for CMY color error diffusion. The algorithm, which we call semi- vector quantization, has a low computational complexity and a high stability as similar to scalar error diffusion, but yields superb quality images close to those generated from vector error diffusion.