Quantum chemistry calculations on a quantum computer frequently suffer from symmetry breaking: the situation when a state of assumed spin and number of electrons is contaminated with contributions of undesired symmetry. The situation may even culminate in convergence to a state of completely unexpected symmetry, e.g. that of for a neutral species while a cation was expected. Previously, the constrained variational quantum eigensolver (CVQE) approach was proposed to alleviate this problem [Ryabinkin et al. (2018), J. Chem. Theory Comput. DOI:10.1021/acs.jctc.8b00943] here we analyze alternative, more robust solutions. In particular, we investigate how symmetry information can be incorporated directly into qubit Hamiltonians. We identify three essentially different techniques, the symmetry projection, spectral shift, and spectral reflection methods, which are all capable of solving the problem albeit at different computational cost, measured as the length of the resulting qubit operators. On the examples of LiH and HO molecules we show that the spectral shift method, which is equivalent to penalizing states of wrong symmetry, is the most efficient, followed by spectral reflection, and symmetry projection.