I've not read the paper fully but it does look an interesting piece of work. I did notice that the controls were aged and IQ matched which doesn't suggest they we looking to differentially diagnose Autism from other disorders which is likely to be the more useful clinical technique. Also, the subjects were adult and I would expect most diagnosis to be performed in a much younger age group. This isn't the first attempt to apply fMRI techniques to the understanding of autism, plenty of prior work here (http://goo.gl/oW41Vm).

Earlier in my career I did some fMRI and, whilst an interesting technique, I've never been convinced that it quite delivers on the claims. Firstly, you aren't looking at an electrical signal but rather changes in blood flow which are (reasonably) assumed to be coupled with areas of increased neural activity. This felt to me a bit like trying to debug a computer program by pointing a heat sensitive camera at your motherboard.

Also all fMRI experiments are characterised by a block design - periods of activation and inactivation - and the response is expected to correlate with these in some way. However, the brain is pretty busy and thus how does one select a baseline; what about the parts of the brain that just stay on?