It is easier to indicate the ink color of a color-neutral noun when it is presented in the color in which it has frequently been shown before, relative to print colors in which it has been shown less often. This phenomenon is known as color-word contingency learning. It remains unclear whether participants actually learn semantic (word-color) associations and/or response (word-button) associations. We present a novel variant of the paradigm that can disentangle semantic and response learning, because word-color and word-button associations are manipulated independently. In four experiments, each involving four daily sessions, pseudowords—such as enas, fatu or imot—were probabilistically associated with either a particular color, a particular response-button position, or both. Neutral trials without color-pseudoword association were also included, and participants’ awareness of the contingencies was manipulated. The data showed no influence of explicit contingency awareness, but clear )