This paper experimentally dissects the preferences for honesty into two components: lying aversion and deception aversion. For a separate identification, we consider two reputation-building environments with a two-dimensional belief domain, where lying without deception occurs in one environment and deception without lying occurs in the other environment as a unique equilibrium phenomenon. The choice data collected in the lab enable us to differentiate between individuals' aversion to making statements that are literally untrue (lying aversion) and their aversion to statements intended to manipulate others' beliefs (deception aversion), controlling for inference errors.