You don't say what errors you get.
You could get problems if you are in a region where y is near zero. Away from that region, here's an example where I compute the mean of a sample from the distribution, then verify it is close to the value I get by integrating over a finite but large region:
>> mu = [10 10];
>> sigma = [1 1;1 2];
>> xy = mvnrnd(mu,sigma,1000);
>> mean(xy(:,1)./xy(:,2))
ans =
1.0103
>> dblquad(@(x,y)(x(:)./y).*mvnpdf([x(:) repmat(y,numel(x),1)],mu,sigma),1,20,1,20)
ans =
1.0107
