Alternative to McNemar's Test

Hello all,

I want to run the McNemar's Test for matched binary data, however my sample is small (n=10 pairs), and the normal approximation cannot hold. What should I do ?

Looking at McNemars test in Wikipedia it says that “The binomial 'sign test' gives an exact test for the McNemar's test”, so looking at the sign test suggest to test with the binomial distribution. That can be done easily in R. With Wikipedia's example:

 1 - pbinom(15,15+7,0.5)
[1] 0.0262394