We use X-ray phase contrast imaging to characterize the inner surface roughness of deuterium-tritium (D-T) ice layers in capsules for future ignition experiments. It is therefore important to quantify how well the X-ray data correlate with the actual ice roughness. We benchmarked the accuracy of our system using surrogates with fabricated roughness characterized with high precision standard techniques. Cylindrical surrogates with azimuthally uniform sinusoidal perturbations with 100-m period and 1-m amplitude demonstrated 0.02-m accuracy limited by the resolution of the imager and the source size of our phase contrast system. Spherical surrogates with random roughness close to that required for the D-T ice for a successful ignition experiment were used to correlate the actual surface roughness to that obtained from the X-ray measurements. We compare first the average power spectra of individual measurements. The accuracy mode number limits of the X-ray phase contrast system benchmarked against surface characterization performed by atomic force microscopy are 60 and 90 for surrogates smoother and rougher than the required roughness for the ice. These agreement mode number limits are about 100 when comparing matching individual measurements. We will discuss the implications for interpreting D-T ice roughness data derived from phase contrast X-ray imaging.