Leaf gas-exchange models are increasingly used to reconstruct ancient atmospheric carbon dioxide (CO2) concentrations. One of these widely used models, the Franks model, requires stomatal size (guard cell width and either guard cell length or pore length), whole-leaf stomatal density, and bulk-leaf carbon isotope composition (delta C-13) from plant fossils. However, natural variations of these parameters within and across plant leaves have not been assessed closely, hindering the application of this model and the evaluation of its associated uncertainties. Here we investigate the range of variations of these parameters, and evaluate their impact on the output of the Franks model in three conifers (Metasequoia, Sequoia, and Taxodium). We introduce a modified cleared leaf method that allows accurate measurements of stomatal size. We show that among the stomatal size parameters, pore length is the most variable. Whole-leaf stomatal density can be accurately estimated in a representative area in the middle portion of a leaf. Variations of delta C-13 values are only slightly above analytical errors within a leaf and between leaves from a branchlet, but a similar to 1 parts per thousand negative shift of delta C-13 during early decay of Metasequoia leaf tissues was observed. Our measured ranges in pore length and whole-leaf stomatal density have the biggest influence on model estimated CO2. To improve model performance, we recommend (1) the use of our modified cleared leaf method to acquire accurate stomatal size and whole-leaf stomatal density measurements from the middle portion of a leaf located at the middle portion of a branchlet; (2) scaling pore length from guard cell length; and (3) a systematic correction of carbon isotope fractionation may be applicable if information regarding tissue decay and fossil preservation is available. We tested our recommendations by reconstructing CO2 from both extant and fossil materials. Franks model-derived CO2 based upon modern leaves collected in 2004 and 2020 (346 and 416 ppm) are close to their targets (378 and 414 ppm) whereas stomatal frequency methods substantially underestimate (285 and 341 ppm). Reconstructed CO2 from the middle Miocene Clarkia deposit (505 and 507 ppm for Metasequoia and Taxodium) are comparable with published results. We conclude that an improved cleared leaf method for accurate measurements of key stomatal parameters and a statistically-informed stomatal counting strategy will improve the performance of the Franks model for reconstructing CO2 using these conifers with wide distributions of fossil records in the Northern Hemisphere since the Cretaceous.