ABSTRACT Despite the rich observational results on interstellar magnetic fields in star-forming regions, it is still unclear how dynamically significant the magnetic fields are at varying physical scales, because direct measurement of the field strength is observationally difficult. The Davis–Chandrasekhar–Fermi (DCF) method has been the most commonly used method to estimate the magnetic field strength from polarization data. It is based on the assumption that gas turbulent motion is the driving source of field distortion via linear Alfvén waves. In this work, using MHD simulations of star-forming clouds, we test the validity of the assumption underlying the DCF method by examining its accuracy in the real 3D space. Our results suggest that the DCF relation between turbulent kinetic energy and magnetic energy fluctuation should be treated as a statistical result instead of a local property. We then develop and investigate several modifications to the original DCF method using synthetic observations, and propose new recipes to improve the accuracy of DCF-derived magnetic field strength. We further note that the biggest uncertainty in the DCF analysis may come from the linewidth measurement instead of the polarization observation, especially since the line-of-sight gas velocity can be used to estimate the gas volume density, another critical parameter in the DCF method.