I should clarify that I do not really believe there's a "right" number for autism prevalence. It's sort of like asking what is the prevalence of "short" or what is the prevalence of "light skinned". It depends on the boundary, and the boundary in autism is subjective. I believe it is somewhat subjective even when the best diagnostic tools are used. The standard deviation boundary of mental retardation is an analogous boundary, yet probably a more objective one.
Nevertheless, it appears that when ASD is screened thoroughly in a population, or when there's a lot of awareness and good ascertainment, prevalence is found to be closer to 1%. This is not new. The following is what Lorna Wing and David Potter said on the subject as early as 1999.
Because we concentrated on the children with learning disabilities (IQ under 70) we saw very few with the pattern described by Asperger. We had to wait for the study by Christopher Gillberg in Gothenberg to find out how many children with IQ of 70 and above were also in the autistic spectrum. As described above, combining the results of these two studies gave an overall prevalence rate for the whole autistic spectrum, including those with the most subtle manifestations, of 91 per 10,000 - nearly 1% of the general population.
Kadesjö et al (1999) report a study in Karlstad, a Swedish town. Although this was small scale it was very intensive (over 50% of the 7 year old children seen and assessed personally by the first author). The study found a prevalence for all autistic spectrum disorders for all levels of IQ, of 1.21%!!! Children were followed up four years later and had the diagnoses confirmed.
I believe Kadesjö's personal assessment of 50% of 7 year olds in Karlstad is key to that finding.
More recently, Baird et al. (2006) found a prevalence of 1.16% in South Thames, UK. In this study the authors, again, looked for children with a possibly undiagnosed ASD. Note that children in this study were born between 1990 and 1991, before the bulk of what is referred to as the "autism epidemic" got started.
Finally, we have Fombonne et al. (2006) which found a prevalence of 1.076% among Kindergarten children (born in 1998) in Montreal, Canada. While there are fluctuations in the prevalence by birth year cohort in this study, there are no indications that prevalence is about to level off just yet.
Additional supporting evidence comes from Posserud et al. (2006) out of Bergen, Norway. The authors found that 2.7% of 7-9 year olds were high scorers in the ASSQ.
In adults, prevalence of ASD can also be 1% easily considering that in studies of the properties of the AQ Test about 2-3% of adults score above 32-33, the cut-off at which an ASD is considered possible.