Surface roughness of technical surfaces is an important parameter in, for example, quality control. Speckle interferometry (SI) is a powerful tool for acquiring information about a surface under test. Recent investigations concentrated on deriving analytical functions to describe the dependency of surface roughness on SI, assuming normally distributed surface roughness. The common approach is to use the correlation of the standard deviation of height distribution σ and single speckle-related parameters like fringe visibility V or spectral speckle correlation C to estimate surface roughness. Furthermore, roughness cannot be described clearly using only one parameter (e.g., Ra ), which makes it often necessary to estimate more roughness parameters. A new approach in roughness measurement using SI is presented. A multivariate data analysis for generating a regression model is employed, which may include many speckle-related parameters on one hand and offers the possibility to acquire different roughness parameters on the other hand. Finally, the regression models created for four exemplary roughness parameters Rq , R3z , Rp , and Mr1 are discussed and the accuracy in the prediction of these parameters is indicated.