A systematic screening procedure for small contributions in the incremental expansion of the correlation energy is presented. The performance of the proposed scheme is checked for the calculation of intermolecular interactions in realistic test systems as large as a guanine - cytosine base pair. It is found that the computational cost for the incremental expansion can be reduced considerably without significant loss of accuracy. Typically, the errors of the systems investigated here amount to < 3.4, 0.22, and 0.06% for second-, third-, and fourthorder expansions, respectively. For almost all cases, the error in the total correlation energy can be kept below 1 kcal/mol with respect to the canonical CCSD result if the incremental series is truncated in a proper way.