This paper focuses on the use of data mining as an enabler of the new audit risk. In general, data mining techniques which are classified by nature offer the capacity to consider multipletypes of evidence and can assist auditors in assessing risk, issuing opinion, and making judgement. Especially, this study investigates the efficiency of ensemble techniques in developing models for estimating audit risk, as opposed to models developed with single techniques including logit regression, decision tree, and neural network analyses. The sample consists of 302 financial statements that received a qualified audit opinion and accounting disclosure reviews and investigations over the period 2000-2010. The input variables are composed of four sectors with 75 predictors including suspicious journal entries,incentive factors and red-flags, properties of firms, and audit firms' characteristics and audit risk. And then, comparing the performance of a single and ensemble techniques in detecting fraud under different assumptions of misclassification costs and ratios of fraudulent firms to non-fraudulent firms, the single techniques used are logit regression, decision tree and neural networks and the ensemble techniques used are bagging, boosting, and combined ensemble. In each case of the sampling, ensemble techniques are revealed more efficient, in terms of average classification accuracy and under different assumptions of misclassification cost than the single methods. Also, the results of the industry-specific analyses are equal. In conclusion, this study reports the positive results of an analyses examining the effects of audit risk assessment instructions on the effectiveness of data mining. Especially, ensemble techniques are better then single technique. And these results have implications for the audit risk assessment of auditors to do the tasks and the structuring of these tasks.