Classifications have to meet a variety of purposes. Clinical and research needs are different and there is much to be said for separate clinical and research schemes. Care is needed to ensure that classifications provide an appropriate medium for teaching about diagnosis and do not cause difficulties when used as a "passport" to resources. Principles of classification are considered in relation to the need to take course, as well as symptomatology, into account, and with respect to the neuropsychiatric interface. The value of a multiaxial approach is noted. The pros and cons of autism and pervasive developmental disorders (PDD) as an overall descriptive term, of lumping or splitting, and of different choices with respect to PDD subcategories are discussed.