Since the existing crowdsourcing model could not detect the malicious behavior in the crowdsourcing system quickly and efficiently, a reputation model based on Dempster-Shafer theory, called DS_CQC (Dempster/Shafer Crowdsoucing Quality Control), was proposed to apply to the crowdsourcing quality control. Firstly, the sustainable credible evidence sequence and sustained incredible evidence sequence based on time-window were obtained. Secondly, the original D-S evidence theory was improved through three aspects including importance of evidence, relationship of evidence and reliability of witness, and the new basic probability function was acquired. Finally, evidence sequence was fused by using the improved D-S evidence theory and then the direct reputation, indirect reputation and comprehensive reputation were computed. The incentive mechanism based on reputation was used to encourage people to participate in crowdsourcing actively and submit a higher quality crowd, while the malicious workers were suppressed. Experiments on simulation and real crowd data were conducted, and compared to the trust model of probability, the detection of malicious behavior in the crowdsourcing system of DS_CQC increased by 50% in speed and 3.1% in efficiency at least. The result proves that the DS_CQC has the high anti-attacking capability.