The problem with saying an exact number is that PCR is the gold standard, so it's hard to compare a gold standard with a gold standard.
Some of the early Chinese data suggested 70% to 80% accuracy. Those were really poorly constructed tests, with lots of different reagents and lots of different parts. They were not great. Better accuracy is probably at 90% to 95%. Again, it depends on where you're getting them and the context of their infections. Very early or very late, you can still miss people.
Looking at data from the Ontario public health labs, when people are tested, in the few tests that are done serially—I think there were about 8,000 cases, or something like that, where people were tested twice, which we often do in hospitals when we are very suspicious—there were about 2.5% that flipped over from negative to positive on the next test, so you do miss people.
Again, this is the gold standard, so it's hard to say what's better than this in that sense.