I did the MARS Wisdom Panel one. Although it wasn't accurate, they are one of the best companies when it comes to customer service. If you can catch it on sale and look at it more as something fun to do instead of expecting real results, I'd definitely suggest that one. It's got the best reviews of the reasonably priced ones and they provide so much information. If you think your dog may include one type of breed, you can list it and they'll test the DNA against other samples. Of course, the breed that vets, breeders, trainers, rescue workers, and even a military dog trainer say that my dog primarily is didn't come back at all in the results.
Take it with a grain of salt and have fun with the results is my opinion of it.
Mars company, the same company that makes candy, put a patent on the idea of doing a genetic test to find what breed a dog is made up of, so regardless of what test you buy, in the end all the info goes to Mars and gets processed through their computer.
The process looks at 'nonsense' bits of DNA. A lot of doggie 'features' like coat color are decided by a single pair of genes, so a dog can have a coat like breed X, but be many generations removed from X, so looking at the 'nonsense' section is actually a smart idea.
Flaws in the system are that they examine a handful of dogs from each breed to create their profile. Because of heavy use of popular show sires, even 4 or 6 generations back, your sample dogs may have a shared ancestor and it may pick up a sequence that ancestor passed on to all his progeny that happens to be in all the sample dogs but not present in purebred dogs of the same breed from a radically different line.
Additionally, a pattern they may find in 10 out of 10 of their sample breeds (note I don't know if they get 10 of a purebred to look at to find the 'key bit' or more or less...I actually suspect less) and that same pattern unknown to the testers can be found in 1 out of 10 of a totally unrelated breed. For instance Appenzeller Sennenhund is a rare breed, and has a very disticnct marker and all Appenz have it....but it turns out it turns up in a lot of other breeds, and when there aren't other indicators to help the test steer to the correct breed, it goes by the one indicator it has and pops up dogs as being mixed Appenzellers a lot.
So when the poster above states that wisdom panel frequently mistakes a Border Collie for as Chiwawa I think it's probably something similar.
I used Wisdom Panel. It was correct for my dog, at least it seemed to be. I'm not a geneticist so not sure if it was exactly correct. But, it said my dog was primarily an Australian Kelpie and Setter mix, which is spot on as far as his physical features go. I bought one for my new addition and am waiting for the results. She appears to be a full ACD.