It is essential that the correlation between variables is considered properly when using sampling-based methods. Modeling rainfall events is of great interest because the rainfall is usually the major driving force of hydrosystems. A novel method for generating correlated samples is introduced providing that the marginal distributions of variables as well as their correlations between them are known. The basic idea of the method is to adjust the correlations between samples by rearranging the positions inside marginal samples after each marginal sample is generated according to its distribution. The group method is developed in order to facilitate an efficient generation of correlated samples of large sizes. The theoretical precision associated with the group method is derived. There is a trade off between the computational efficiency of the algorithm and the precision that can be achieved when using different numbers of groups. The method is successfully applied to two cases of rainfall sample generation problems. The effectiveness of the group method is studied. Large group numbers are recommended in practical use as the samples distribute more broadly regardless of computational efficiency.