Saddlepoint approximations for the distribution of the difference of o
rder statistics from a continuous distribution are presented. The appr
oximations are illustrated for the case of the exponential distributio
n where we can find the exact probability density function of the diff
erence of order statistics, and for the case of the noncentral chi-squ
ared distribution which has application to the problem of testing a ti
me series for linearity. In both cases we show that the saddlepoint ap
proximation is much better than the normal approximation, especially f
or realistic sample sizes.