Characterization of gratings with small period-to-wavelength ratios is difficult to perform but is very helpful in improving the fabrication process. We experimentally tested an inverse-scattering method using a neural network on silicon etched gratings. We also characterized the gratings by using two popular microscopic methods. The validity of each method was determined by comparing measured diffracted intensities with calculated ones obtained from measured profiles. An estimation of accuracy and repeatability was deduced from a scan along a grating sample. This method was thus well validated for nondestructive and noninvasive measurements under experimental conditions that were close conditions of actual usage. This method is easy to implement and requires the measurement of only a few diffracted intensities.