Razib Khan debunks the myth “that in 10 percent of cases paternity is misattributed”:
[V]ery high estimates of cuckoldry come from databases of disputed paternity, which are obviously going to be a biased sample. A more thorough survey suggests that there is a wide variation in misattributed paternity across populations.
He points to a German study which estimates nonpaternity in Germany at roughly one percent:
The sample consists of the families of children who require bone marrow transplants. The authors note two important conditions: 1) the details of the results as they might relate to paternity are not divulged, 2) none of the parents refused to be typed. Since susceptibility to childhood cancers are evenly distributed across the population the biases introduced in other surveys presumably do not apply to this situation.