A very common practice in the investigation of suspected sexual abuse is to use anatomically detailed dolls to interview children. The use of such dolls is particularly advocated for very young children. For a doll to be useful, however, children must accept and use it as a representation of themselves. Our previous research on 2- and 3-year-old children's understanding of symbolic objects led us to hypothesize that such very young children might have difficulty understanding and using a doll as a self-representation. In the study reported here, 2-1/2-, 3-, and 4-year-old children played some games with an experimenter, and they were interviewed immediately afterward. The children did, as expected, have difficulty using the doll as a self-representation and mapping from themselves to the doll. As a consequence, they provided more correct information in their direct (verbal and nonverbal) responses to the interviewer's questions than they demonstrated on the doll. Implications and limitations of this research are discussed with respect to interviewing young children.