Ultrasonic inspection of ordinary samples with more or less rough surfaces is an everyday problem in industrial NDE. Contact techniques require flat or other regular (e. g., cylindrical) surfaces of negligible roughness with respect to the acoustic wavelength. Immersion techniques are less susceptible to surface topography, but they still require that the surface radius be larger than the beam diameter and the surface roughness be comparable or less than the wavelength in the immersion fluid. This difference is due to the fact that in immersion inspection surface irregularities do not significantly reduce the energy transmission into the specimen but rather randomize the field through incoherent scattering. Figure 1 shows the schematic diagram of ultrasonic inspection of a rough specimen by the immersion method. The probability of detection of a given flaw is ultimately limited by the signal-to-noise ratio produced at the receiver. The flaw signal results from coherent reflection from a single, relatively large and strong scatterer. In comparison, the noise is incoherent scattering from a large number of randomly distributed, relatively small and weak scatterers such as material inhomogeneities or geometrical irregularities. Surface roughness can substantially reduce the signal-to-noise ratio with respect to an otherwise similar smooth sample. First, surface roughness attenuates the coherent flaw signal much more than the incoherent material noise [1,2]. Second, surface roughness increases the overall noise level by adding another incoherent component to the material noise. This paper discusses the adverse effect of the excess surface noise on ultrasonic flaw detection in rough samples.