The noise covariance matrix, whose diagonal square root is commonly referred to as radiometric noise (NESR), is one of the most important characteristics of hyperspectral infrared sounders measurements. It is used in spectral data inversion and in estimating the atmospheric state vector. This paper presents new results of noise covariance matrix characterization in measurements of the IKFS-2 infrared Fourier transform spectrometer, which has been successfully operating on board the Meteor-M No. 2 spacecraft for more than 6 years. The main factors leading to the interchannel noise correlation are considered. They are associated both with the properties of noise in the interferograms measured by the instrument and the specifics of the initial processing procedure. The noise covariance matrix in the IKFS-2 output spectra has been experimentally estimated in three different ways: (1) from measurements of reference radiation sources, (2) from measured atmospheric spectra, and (3) from the imaginary part of the calibrated atmospheric spectra. The results of the experimental assessment are consistent with the calculations and can be used in the problems of thematic processing and assimilation of IKFS-2 data in numerical weather prediction models of Roshydromet.