Geographically extensive forest inventories, such as the USDA Forest Service’s Forest Inventory and Analysis (FIA) program, contain millions of individual tree growth and mortality records that could be used to develop broad-scale models of forest dynamics. A limitation of inventory data, however, is that individual-level measurements of light (L) and other environmental factors are typically absent. Thus, inventory data alone cannot be used to parameterize mechanistic models of forest dynamics in which individual performance depends on light, water, nutrients, etc. To overcome this limitation, we developed methods to estimate species-specific parameters (θG) relating sapling growth (G) to L using data sets in which G, but not L, is observed for each sapling. Our approach involves: (1) using calibration data that we collected in both eastern and western North America to quantify the probability that saplings receive different amounts of light, conditional on covariates x that can be obtained from inventory data (e.g., sapling crown class and neighborhood crowding); and (2) combining these probability distributions with observed G and x to estimate θG using Bayesian computational methods. Here, we present a test case using a data set in which G, L, and x were observed for saplings of nine species. This test data set allowed us to compare estimates of θG obtained from the standard approach (where G and L are observed for each sapling) to our method (where G and x, but not L, are observed). For all species, estimates of θG obtained from analyses with and without observed L were similar. This suggests that our approach should be useful for estimating light-dependent growth functions from inventory data that lack direct measurements of L. Our approach could be extended to estimate parameters relating sapling mortality to L from inventory data, as well as to deal with uncertainty in other resources (e.g., water or nutrients) or environmental factors (e.g., temperature).