The invention relates to matrix image sensors, and it relates more particularly to a method for correcting the fixed spatial noise of the matrix.The signal (Sij,n) obtained from each pixel (Pij) during a large number of successive images is collected; for each pixel, an approximate value (Mij) of an average of the signal obtained from the pixel during this large number of images is determined; the signal obtained from each pixel is corrected by adding to it the difference between a reference average value (M0) and the approximate average value. The approximate average is obtained by determining high and low limit values taken by the signal obtained from the pixel during the large number of images, and by calculating a median between these high and low limit values.