The figure usually given for risk of a blow job on an HIV+ man is 0-0.04% (0 to 4 in 10,000). The high end comes from one study that looked at the behaviour of gay and bisexual men in three American cities over an extended period. By seeing who became infected and who didn't, and looking at what behaviour they reported, they came up with the figure... but the 'confidence interval' went a little bit higher and way lower.
What's that mean? When you look at a sample of something, you don't know if you happen to have got the right figure for the whole group. It could
be the sample you looked at was odd, for example. So there exist a variety of mathematical tests that tell you things like 'your answer is 0.04%, and there's a 95% chance that this is about right, and a 5% chance that it's absolute crap'. Less than 95%, and everyone goes 'it could well be crap', much more than 95%, such as 99% or the Holy Grail of 99.9% and above, and you can make a bigger fuss about what you've found.
There are other tests which say, 'ah, but given this data, any figure between 0.001% and 0.05% would still had a 95% chance of being
"about right"'... and this is what happened here. It's quite possible that 0.04%, while low, is still a big overestimate of the risk, particularly as there are other studies reckoning the risk is almost nil. Given the assumptions they had to make - oral is very, very common when men have sex with each other - I personally reckon it is an overestimate. (If it weren't, they'd be a lot more HIV+ gay men!)
If he doesn't come in your mouth, it's nil or so close to nil that you're more at risk of dying through a heart attack while doing it.