This paper addresses the problem of evaluating noise suppression methods for images produced by synthetic aperture radar (SAR) systems. It is demonstrated that noise in SAR images is multiplicative and exhibits characteristics between Rayleigh and Gaussian distributions. It is shown also that such noise is typically spatially correlated and features an asymmetric spatial spectrum. Our study highlights that the information available about SAR image formation is insufficient to reliably estimate the noise spectrum directly, necessitating its estimation before noise suppression. A test set of images with noise characteristics closely resembling those of real SAR data is proposed. This dataset, called SAR400, enables effective benchmarking of noise suppression methods. Using the proposed test set, several existing methods for suppressing spatially correlated noise are evaluated, and their limitations are analyzed.