The estimation of the chromaticity of an ambient light is relevant to many color-based vision applications, like object/people recognition, video-surveillance and assisted living. In this paper, we propose to estimate the chromaticity of an environmental light by means of a smart low-power color sensor. The sensor is composed by a RGB pixel connected to a microcontroller, embedding signal processing on board. The RGB pixel acquires the color signals in a high dynamic range with low power consumption. The microcontroller selects the signals coming from achromatic surfaces, whose chromaticity is equal to the light chromaticity, the chromaticity of the selected signals are sent to a processor, that returns the most frequent chromaticity as the most probable light chromaticity.