The surface roughness is evaluated by decomposing micro-errors from irregular surface image. The wave of surface is designated as signals including high-frequency, mid-frequency, and low-frequency signal, that denote roughness, waveness and the geometry shape error. Hilbert-Huang transform is a promising revolutionary technique for spectral data analysis, which is used to extract roughness from surface. It overcomes the imprecise result of the traditional surface roughness calculation and avoids the complication of the Wavelet analysis. Theory of Empirical Mode Decomposition (EMD) is given in the paper. This method is initially applied in picking up surface roughness, and it is proved to be of very high efficiency and simplicity.