Merits and limitations of Koopmans' approximation in estimating the vertical and adiabatic ionization potentials are examined by the G2(MP2) procedure in some small molecules involving first and second row atoms. It is shown that Koopmans' theorem and the corresponding orbital energies provide a very crude approximation to the ionization potentials IPs in a quantitative sense. In fact, vertical and adiabatic potentials are given by an interplay of four and six terms, respectively, which are not simply proportional to the orbital energies and usually do not vary in a predictive way. However, it is possible to find satisfactory linear relations between the orbital energies and both types of the ionization potentials for families of very closely related molecules. On the other hand, the many-body effects are important for a quantitative description of the phenomenon and in interpreting changes in IPs between widely different molecules. A serious discrepancy is found between the G2(MP2), G2, and G3 adiabatic ionization potentials ADIP for a series of molecules NHnF3-n (n = 1-3) and the experimental data. This would suggest a significant contribution of the vibrationally excited states of the ion to the experimental ADIP values, particularly since the predicted vertical ionization potentials are in harmony with the corresponding observed data. The influence of the continuum states, which is not taken into account in the theoretical model, cannot be excluded either. Alternatively, it is possible that the correlation energy is not correctly reproduced in these molecules by the G2(MP2), G2 and G3 computational protocols.