Color image quantization is very used in several applications of image processing. In this paper, we present an original approach to define quantizers for color images taking into account human visual system properties. As this approach is entirely based on results of psychophysics experiments, we define it as psychovisual quantization. From experiments, we have constructed a model to define the optimal quantization law in various spatial frequency without introducing any visible degradation. This model include three steps: a component decomposition defining a color space, a sub-band decomposition which emulates the frequency tuning of human visual system receptor fields, and finally an adaptation of the quantization law according to the modification of the visibility of a signal depending of its context due to the masking effect. Obtained model is used to define the optimal quantization laws adapted to any color image.