Let X1,...,Xn be a random sample of size n from a DFR distribution and let Xi:n) denote its ith order statistic. It is shown that for i < j, X(i:n) is less dispersed than X(i:n). Also, if Xi's are independent DFR random variables, but not necessarily identical, then X(l:n) is less dispersed than X(j:n), for j> 1.