The problem of negative electronic populations and of occupation numbers greater than 2 has plagued Mulliken Population Analysis since the very beginning. Through the analysis of three model molecular systems, several basis sets and the relevant literature, we conclude that there is not enough evidence to assign the origin of these errors to the self-consistent scheme, to Mulliken's partition, to the basis set structure or to a combination of these. As Mulliken Population Analysis is still widely used, we have developed an empirical method to eliminate negative electronic Populations and occupation numbers greater than 2. This method can be used for any partition of the electron density (not only Mulliken's), for any basis set and for any LCAO-MO methodology (semiempirical or ab initio). Finally, the method does not produce any change in the original atomic net charges.