In the last decades autism diagnosis has shown great differences in prevalence studies, apparently related to methods applied in the evaluation of children with relational and\or communicative difficulties. In the present study a literature evaluation is conducted, suggesting the advantages of an extensive and accurate clinical evaluation of the child with his family, possibly supported by tests like CARS, convergent with this approach and, in contrast, risks related to the use of semi-structured tests. Some of the last-mentioned tests have been considered, decades ago, as `golden standard’ for diagnosis, but this was probably an unfortunate illusion, which may have favoured many inappropriate diagnoses. The history of the concept of autism is described, split between the suggestion of dealing with `one’ condition and, alternatively, with a behaviour related to different syndromes and disorders. Various examples on positive evolution of this condition are reported, supporting the view of autism as an abnormal behaviour, usually comorbid with other conditions, in some cases curable and amenable to correction up to the removal of the diagnosis.