This paper presents a technique for measuring surface roughness (Ra), using photometric stereo method. The method utilizes three or more images of the measured surface from the same view point, taken sequentially under different illumination conditions. The scenes captured by the camera were reconstructed by the least square method to obtain surface normal from surface topography. Three-dimensional surface textural patterns were recovered from those surfaces normal. The system was validated with three standard surface roughness specimens. The Ra calculated from the recovered surface was compared with the nominal value of the standard surface roughness. The best coefficient of the slant angles can also be determined through the linear regression. The experimental results indicate that the technique can be applied to well recover standard surface roughness.