Hypothetical bias in stated-preference methods appears sometimes to be very large, and other times non-existent. This is here largely explained by a model where people derive utility from a positive self-image associated with morally commendable behavior.