Guidelines for revisions to D.S.M.-5 asked those proposing changes to consider potential harms to patients. This has been controversial. This paper argues that seeking to do no harm is appropriate when constructing a scientific classification scheme. I show that in many cases considerations of harm avoidance can play a role in influencing the design of the D.S.M. without the pursuit of scientific knowledge being compromised. I then turn to cases where compromise is required. Through a consideration of cases where lying is required to save lives I suggest that it is widely accepted that harm-avoidance can trump truth-telling. In extreme circumstances a psychiatric classification might thus set out to mislead in order to prevent harm.