The accuracy of measurement of the thickness of uniform thin films on solid substrates by null ellipsometry is severely limited when the substrate is rough. It is impossible to separate these two effects experimentally with the null ellipsometer, and there is no theoretical basis or generally used model available to separate these effects. Thus, a dual rotating‐compensator Mueller matrix ellipsometer has been constructed to carry out film thickness measurements on rough substrates. Measurements were made on a set of specially prepared specimens of 8630 steel, roughened by grit blasting with aluminum oxide. Grit sizes and blasting pressures were varied to produce 11 different roughness values ranging from 0.01 to 1.295 &mgr;mRa, as measured with a stylus tracer device. Upon each of the 11 roughness groups, films of magnesium fluoride were overlaid to thicknesses of 89, 180, 254, and 315 nm. One set of specimens was left uncoated. Experimental results for film thickness measurements on rough surfaces matched the ideal (for smooth surfaces) form well for roughnesses up to 0.13 &mgr;mRa, at most angles of incidence. For rougher specimens, significant deviations in results were observed for all but the largest angles of incidence. The nonideal data were attributed to the cross‐polarization effects of surface geometry, and apparent depolarization. The resolution of thickness measurements was 1 nm for polished specimens, and decreased continuously to 10 nm for the roughest specimens examined.