The Family Smoking Prevention and Tobacco Control Act of 2009 (the Act) mandated that FDA assume regulatory authority for tobacco products. A key provision of the Act states that FDA shall grant a modified risk tobacco product (MRTP) order only if an applicant has demonstrated, among other conditions, that the new tobacco product will result in reduced harm with no concomitant increase in risk to the population as a whole. This implies that both the intended, beneficial consequences and the potential for unintended, harmful consequences must be considered. Because the context for such evaluations involves predicting the potential effect of a new exposure, statistical models are necessary. By comparing counterfactual scenarios where cigarettes and the MRTP are available and a base case where cigarettes are used exclusively, statistical models estimate changes in population mortality which might result from projected changes in exposure patterns. Model results can be used to predict the magnitude, and thus likelihood, of changes in exposure patterns needed to produce population benefit or harm. To be responsive to regulatory requirements, models must be based on a scientifically valid conceptual framework. They must be flexible, allowing users to input and modify scenarios easily, and all required assumptions must be clearly documented. Models should be calibrated and results should be validated against existing population data. We examined the specific model requirements identifiable from the Act, and assessed each of several published models proposed to satisfy regulatory requirements. Most are not directly relevant to regulatory requirements or suffer from conceptual shortcomings. In particular, models that attempt to project the effects of introducing an MRTP to a cross-sectional, mixed age population require untenable assumptions and may produce biased results due to missing exposure histories and incomplete follow-up.