A method is presented which a relates a parameter describing rock surface roughness (a gradient statistic) to the frictional strength of discontinuities (the friction angle). The various statistics used to describe rock surface roughness are reviewed together with their interrelations. The implicit “filtering” involved in the sampling process and the associated effects on the prediction of φ are discussed. A surface contact model is adopted which assumes Gaussian probability distribution functions for surface height, gradient and curvature. It further assumes that contact involves truncation of the surface but no distortion and that contact “spots” behave in a predominantly elastic manner. The well-known dependence of the rms (root mean square) gradient statistic (z 2) on the sampling interval (dx) is overcome by using the average asperity gradient, which can be shown to be largely independent of dx for many rock surfaces. This value is shown to be a function of the “effective” contact spot radius (r) and the dimensionless normal pressure (p). The rms gradient z 2 (r) for a sampling interval equal to the “effective” contact spot radius is shown to be a simple function of the average asperity gradient (Z 1/ xD) . This latter parameters proves to be a useful predictor of friction angle from surface roughness statistics alone. The equations developed provide some support from surface roughness considerations for Barton's formula [2] which is purely empirical relation between friction angle and dimensionless normal pressure ratio. The limitations of Barton's formula are discussed and alternative prediction procedures suggested. An attempt is made to validate a method for estimating friction angle from surface roughness characteristics alone for any given dimensionless normal pressure ratio (p). A calibration curve is constructed from published data and used to predict φ for a totally independent dataset. The results are encouraging but also highlight inadequacies in the available experimental database. A second test of the prediction procedure is made by estimating the tau-sigma relations for surfaces using published shear strength and surface roughness profile data. Agreement between observed and predicted shear strength envelopes is within the expected error bounds.