Based on Kirchhoff approximation method, a numerical computation model of Mueller matrix full angle distribution is constructed herein, and Mueller matrix distributions of random rough surfaces of metals and dielectrics are calculated. Results show that there are obvious difference between Mueller matrix distributions of metals and dielectrics; six components of the Mueller matrix of dielectric surfaces are close to 0, whereas those of metal surfaces are non-zero; additionally, the difference is independent of the surface roughness. This difference can act as the robust criterion to distinguish metal and dielectric targets, and can provide new tools in target detection and recognition.