Suppose sufficient balls are placed at random in n boxes up until the first point at which only x boxes are empty, and that number of balls is r: then the probability that exactly r balls are used for given x and n is S2(r-1,n-x-1) * n!/(x!*n^r) where S2(a,b) are Stirling numbers of the second kind and can be calculated with S2(a,b) = b*S2(a-1,b) + S2(a-1,b-1) starting with S2(0,0)=1 and S2(a,0)=0 if a>0. In this case x=0 and n=36 so the probability becomes S2(r-1,35) * 36!/36^r This is not difficult to calculate for individual values of r except that some of the numbers are very large. As far as I can tell, the median is r=142, the 99% cumulative probability comes with r=291, and the mode is at r=129. These compare with the easier to calculate mean of Sum_{1<=j<=36} 36/j = 150.284... [ in general Sum_{x+1<=j<=n} n/j ] and standard deviation of sqrt(Sum_{1<=j<=36} (36/j)*(36/j-1)) = 44.114... [ in general sqrt(Sum_{x+1<=j<=n} (n/j)*(n/j-1)) ] so the 99% point comes almost 3.2 standard deviations away from the mean, which is plausible. As far as I can tell ... the 99% cumulative probability comes with r=291 A basic result: Suppose r balls are placed in n boxes at random and x boxes are empty, then the probability that x boxes are empty for given r and n is Prob(X=x|r,n,C) = S2(r,n-x) * n! / (x!*n^r) Proof by induction: a) True for r=0 since if x=n: Prob(X=n|r,n,C) = 1 = S2(0,0) * n! / (n!*n^0) and if x