We examine to what extent a person’s moral preferences can be inferred from observing their choices, for instance via experiments, and in particular, how one should interpret certain behaviors that appear deontologically motivated. Comparing the performance of the direct elicitation (DE) and multiple-price list (MPL) mechanisms, we characterize in each case how (social or self) image motives inflate the extent to which agents behave prosocially. More surprisingly, this signaling bias is shown to depend on the elicitation method, both per se and interacted with the level of visibility: it is greater under DE for low reputation concerns, and greater under MPL when they become high enough. We then test the model’s predictions in an experiment in which nearly 700 subjects choose between money for themselves and implementing a 350e donation that will, in expectation, save one human life. Interacting the elicitation method with the decision’s level of visibility and salience, we find the key crossing effect predicted by the model. We also show theoretically that certain Kantian postures, turning down all prices in the offered range, easily emerge under MPL when reputation becomes important enough.