Coverage-guided fuzzing has become one of the most effective ways of vulnerability detection. The widely used edge coverage is a kind of control flow information. However this feedback information is too coarse when detecting taint-style vulnerabilities. A large number of taint-independent seeds are added to the queue, and the number of taint-related seeds converges prematurely, which leads to the loss of evolutionary direction of fuzzing and unable to efficiently test the information flow between source and sink. Firstly, the reasons why the existing feedback mechanism is not efficient enough in detecting taint style vulnerabilities are analyzed. Secondly, TaintPoint, a fuzzer dedicated to taint style vulnerability detection, is proposed. TaintPoint adds live taint as data flow information on the basis of control flow traces to form the live trace as coverage feedback, and the live trace is used to improve the existing method in the instrumentation, seed filtering, selection, and mutation stages respectively. Experimental results on UAFBench show that the efficiency, output, and speed of TaintPoint in detecting taint-style vulnerabilities surpass the industry-leading general-purpose fuzzer AFL++ and directed fuzzer AFLGO. In addition, four vulnerabilities arefound and confirmed on two open source projects.