In this paper we show that the conditional distribution of perturbed chi-square risks can be approximated by certain distributions including the Gaussian distributions. Our results are of interest for conditional extreme value models and multivariate extremes as shown in three applications.