Turbulent flow over a rough surface produces sound that radiates outside the near wall region. This noise source is often at a lower level than the noise created by edges and bluff body flows, but for applications with large surface area to perimeter ratios at low Mach number, this noise source can have considerable levels. In the first part of this dissertation, a detailed study is made of the ability of the Glegg & Devenport (2009) scattering theory to predict roughness noise. To this end, comparisons are made with measurements from cuboidal and hemispherical roughness with roughness Reynolds numbers, hu_Ï /ν, ranging from 24 to 197 and roughness height to boundary layer thickness ratios of 5 to 18. Their theory is shown to work very accurately to predict the noise from surfaces with large roughness Reynolds numbers, but for cases with highly inhomogeneous wall pressure fields, differences grow between estimation and measurement. For these surfaces, the absolute levels were underpredicted but the spectral shape of the measurement was correctly determined indicating that the relationship of the radiated noise with the wavenumber wall pressure spectrum and roughness geometry appears to remain relatively unchanged. In the second part of this dissertation, delay and sum beamforming and least-squares analyses were used to examine roughness noise recorded by a 36-sensor linear microphone array. These methods were employed to estimate the variation of source strengths through short fetches of large hemispherical and cuboidal element roughness. The analyses show that the lead rows of the fetches produced the greatest streamwise and spanwise noise radiation. The least-squares analysis confirmed the presence of streamwise and spanwise aligned dipoles emanating from each roughness element as suggested by the LES of Yang & Wang (2011). The least-squares calculated source strengths show that the streamwise aligned dipole is always stronger than that of the spanwise dipole, but the relative magnitude of the difference varies with frequency.