Alternative to two-sample t-test when proportions are SMALL (<0.001)


I'm looking for an alternative to the two-sample (independent) t-test for equality of proportions.

The two-sample t-statistic only holds under the condition that np>5 and n(1-p)>5. If the true p1 and p2 are sufficiently small, and the sample size not sufficiently large enough, then these two conditions can fail. I'm dealing with true probabilities in the range of 0.0001 to 0.0009.

What alternative to the t-stat can I use in this case? The t-stat may be an approximation, but I'm looking for something more precise. The data are not paired. Mann-Whitney?

Thanks for any advice you may have.