Projections of future changes in land carbon (C) storage using biogeochemical models depend on accurately modeling the interactions between the C and nitrogen (N) cycles. Here, we present a framework for analyzing N limitation in global biogeochemical models to explore how C-N interactions of current models compare to field observations, identify the processes causing model divergence, and identify future observation and experiment needs. We used a set of N-fertilization simulations from two global biogeochemical models (CLM-CN and O-CN) that use different approaches to modeling C-N interactions. On the global scale, net primary productivity (NPP) in the CLM-CN model was substantially more responsive to N fertilization than in the O-CN model. The most striking difference between the two models occurred for humid tropical forests, where the CLM-CN simulated a 62% increase in NPP at high N addition levels (30 g N m−2 yr−1), while the O-CN predicted a 2% decrease in NPP due to N fertilization increasing plant respiration more than photosynthesis. Across 35 temperate and boreal forest sites with field N-fertilization experiments, we show that the CLM-CN simulated a 46% increase in aboveground NPP in response to N, which exceeded the observed increase of 25%. In contrast, the O-CN only simulated a 6% increase in aboveground NPP at the N-fertilization sites. Despite the small response of NPP to N fertilization, the O-CN model accurately simulated ecosystem retention of N and the fate of added N to vegetation when compared to empirical 15N tracer application studies. In contrast, the CLM-CN predicted lower total ecosystem N retention and partitioned more losses to volatilization than estimated from observed N budgets of small catchments. These results point to the need for model improvements in both models in order to enhance the accuracy with which global C-N cycle feedbacks are simulated.