AbstractIn recent years, significant progress has been made in fast radiative transfer model (RTM) simulation of daytime nonlocal thermodynamic equilibrium (NLTE) emission. However, NLTE remains as one important reason that prevents the hyperspectral shortwave infrared (SWIR) radiance observations from being assimilated into numerical weather prediction (NWP) models. To better understand the limitations of existing RTM‐based NLTE simulation, this study introduces a new statistical method, called Spectral Correlations to Estimate Non‐Local Thermal Equilibrium (SCENTE), to estimate the NLTE radiances in the Cross‐track Infrared Sounder (CrIS) SWIR radiance observations. SCENTE is applied to four typical season days, including spring equinox, summer solstice, fall equinox, and winter solstice. By analyzing calculation/background minus observation (BMO) of CrIS SWIR brightness temperature (BT), results show that SCENTE characterizes the NLTE well with standard deviation of differences (STD) comparable to observation noise for both daytime and nighttime, while the community RTM (CRTM) has substantially larger STD at night, mainly due to the lack of daytime NLTE just beyond the day/night terminator and the lack of aurora‐related NLTE. Detailed investigation of the biases of BMO shows that CRTM underestimates daytime SWIR NLTE effects by 0.76 K, while SCENTE overestimates SWIR NLTE effects by 0.70 K. The overestimation is because SCENTE uses CRTM‐simulated SWIR local thermodynamic equilibrium (LTE) radiances in the training, which is underestimated by 0.70 K in BT. SCENTE, complementary to RTM‐based simulations, can be used for quality control of SWIR radiances for assimilation and retrieval of atmospheric soundings.