Decisions are virtually always made under some uncertainty, and they are accompanied by a subjective feeling of confidence, which, statistically, should reflect the probability of being correct. However, several motivational biases have been identified, chiefly overconfidence and a sensitivity to contextual affect, which are resistant to learning. We investigated different aspects of these biases in decisions based on learnt subjective preferences and perceptual information, manipulating monetary incentives and decision difficulty. We found that confidence biases in decisions about subjective preferences can be partially explained by confidence overweighting the learnt value of the chosen option, and that the magnitude of confidence biases is predicted by that of learning biases, consistent with the idea that confidence biases are maintained or exaggerated through biased updating of beliefs. We also test the generality of confidence biases and their reliance on aspects of the choice process, using a perceptual decision context. We found that biases are still present, though attenuated, across agency manipulations decoupling choice from confidence judgements. A potential explanation is that incentives exaggerate the objective evidence favoring a covert choice, which may mismatch actual choice in observed or imposed decisions. Based on the general results of this thesis, I propose that confidence judgements are informed by the expected subjective value of the (mentally) chosen option, a mechanism that could parsimoniously explain confidence biases observed across different decision contexts.