Abstract Accurate productivity estimates are essential to assess the overall sustainability of forest resources. Site index (SI) models for loblolly pine (Pinus taeda L.) in plantation and natural forests of the southeastern United States were developed using the Forest Inventory and Analysis (FIA) database. We extracted short (~20 years), unbalanced panel data from the FIA database. Ten different nonlinear models derived from the base models using the algebraic difference approach (ADA) or the generalized algebraic difference approach (GADA) were fitted to the extracted data. The performance of the models was ranked based on a variety of fit and evaluation statistics. The results showed that all top three models were derived using the GADA approach. The best model for loblolly pine plantation and natural forest stands was derived from the Hossfeld model and the Chapman–Richards model, respectively. The best-fitted models for planted forests were also compared with previously developed models. This study demonstrated that base-age invariant and polymorphic SI models could be developed using short panel data extracted from FIA data. The SI models presented here can be used as a height growth model component in forest growth and yield model systems. Study Implications: Improved site index equations for assessing the site quality of loblolly pine plantation and natural stands are now available to stakeholders at the policy, management, and operational levels. Activities such as forest management, restoration, and wildlife management, which require site quality data, will benefit from the new models. Furthermore, the approach of deriving panel data based on Forest Inventory and Analysis data offers information on developing and updating models for other species. Finally, the approach of this study, to use permanent plot measurement data in developing growth and yield models, is cost-effective and time-efficient.