In order to further improve nonlinear auto-adjusting iterative reconstruction t echnique (NAIRT) and reliably apply it to d iagnose actual air flow field,the anti-noise performance of the NAIRT is specially investigated by simulations.A complicated air flow field is simulated with transition object.It is projected by interferometric tomography.The real projections are obtained by Radon transformation.Gaussian random noise signals are simulated to be produced with function RandG.The noise′s relative intensity is expressed with the ratio of signal to noise (S/N).A series of corresponding Gaussian random noise signals are produced with different intensity levels by simple cal culation.The projections containing noises are obtained from the real projections added with noise linea rly.The inverse projection reconstructions are implemented with NAIRT.The reconstructed results are anal yzed in mean square error (MSE) index and compared with the result reconstructed w ith clean projections un der the same conditions.At the noise level of 60dB S/N,NAIRT could reconstruct the model by a decent accuracy.The MSE declins to 1.82×10－5 after about 40iteration cycles.At the nois e level of 10dB S/N,NAIRT could still reconstruct the model by the level of the profile.The MSE is stabilized at 4.91×10－4 after about 40iteration cycles. At the noise level of 100dB S/N,NAIRT could accurately reconstruct the mo del by micro-detail.The MSE declines to 1.26×10－6 at the end of 103iteration cycles.The reconstruct ed result is very close to that reconstructed with clean projections.Both the reconstructed images and MSE analyses demonstrate that NAIRT has wonderful anti-noise pe rformance.