We introduce a new opponent color space, which aims to mimic the color processing in the primate retina. Recent data from physiology and morphology are used to determine the model and its parameters. The input signal in LMS representation is first nonlinearly transformed by a response saturation stage. Then a linear, spatially extended transformation computes the opponent representation. In contrast to other opponent color spaces, our model does not have a luminance component. Our three components are blue-yellow, red-green, and green-red. It is the spatial extension that allows us to distinguish between the last two. Furthermore, we show that we can reverse the transformation and obtain a LMS representation from the opponent encoding.