Propagating Poisson Errors with small statistics

I have generated Monte Carlo Signal and Background samples and placed the signal samples and weighted them according to luminosity and cross section. However in each bin the statistics are rather small (1 or less). I need to integrate the Signal Plus Background and Subtract the Background only component. However when I use Integral and error it just sums up the error sqr(N) in each bin and adds them in quadrature. I assume that this wouldnt be the correct way in the case of small stastics. So I was wondering what would be the best way to get the Error on the Signal Plus Background and Background and then propagate the errors duing the subtraction of the two. Is there a better way to perform this ?