A series of simulation experiments has been carried out in order to compare the effectiveness and efficiency of three different approaches for determining audit risk in complex audit situations. The first approach uses probability theory and the audit risk model to combine two different items of audit evidence from analytical procedures and tests of details. The second approach is based on the belief function framework and uses the Dempster-Shafer rule for combining the two items of evidence. In the third approach, information from analytical procedures is neglected and audit risk is simply determined by tests of details. The results indicate that the belief based audit is always more efficient than a simple one and can be less efficient than the traditional approach. With respect to audit effectiveness, the belief based approach turns out to be best. Furthermore, the results show that the belief based audit is more robust when there are small failures in interpreting information provided by analytical procedures.