Binary decision diagram (BDD) is a successful data structure for storing a set of bitvectors of a given length. The size of this structure depends on the ordering of stored bits. Sifting is a technique for automatic modification of this ordering that can reduce the BDD size. The goal of the thesis is to present current sifting techniques for BDDs, describe the current implementation of sifting in the library CUDD, adopt this implementation to partial BDDs, extend the sifting in CUDD with the possibility to specify the order of some pairs of bits in all considered orderings, and experimentally evaluate the effect of various sifting settings on the performance of the SMT solver Q3B and its version based on partial BDDs.