The superiority of consequent probabilistic methods for handling and combining uncertainty in expert systems, compared with well known rules like the Dempster-Shafer rule or the MYCIN rule, is demonstrated, even in the case of interval valued probability estimates.