Paw lak rough set model was lacking in giving semantic to positive regions, negative regions and boundary regions. The boundary could not make decision again. But three-way decisions gave a new semantic to boundary regions and we can deal with samples in boundary regions. Based on importance of attribute, samples which meet the conditions were delimit to decision region and others would be maintained in the boundary regions in order to reduce false positives when deal with samples in boundary regions. Based on the study of probabilistic rough set model, three-way decisions-theoretic rough set, Bayesian decision-making process and decisions-theoretic rough set model, this paper presents Three-way Decision mix Decision-Theoretic rough set model(TmD). Compared with the new model with Paw lak-three way decisions model, the loss of division of this model is smaller and the result is more reasonable. This model gives the boundary regions semantic which is delaying decisions. Using three-way decisions makes iterative operation for delaying decisions. In the process of iteration, attributes will be ordered based on the importance of the attribute, thus objectively get the attribute of priority being used in the process of iteration. Experimental results show that the model has a smaller decision cost than only using decision-theoretic rough sets and three-way decisions with iterative operation have a higher accuracy when deal with samples in boundary regions. This paper provides a new method for accurate decision-making.