In Web applications, atomicity violations caused by AJAX (Asynchronous JavaScript and XML) generate non-determinism and inconsistency. This paper introduces a dynamical detecting approach for atomicity violations in AJAX. Implemented based on Jalangi, an existing instrumentation framework, our technique monitors the execution of a Web application and tracks the Function Callback Flow to find atomic regions by taint analysis. Next, we build a dynamic event model, so as to precisely record the related DOM elements in call and callback steps with low overhead. Then we develop an event-based algorithm to acquire the results, which are classified into different severity levels (benign or harmful). Finally, we conduct an empirical experiment on a subset of Alexa top-ranked websites. Our tool detects 175 DOM related atomicity violations in which 48.6% of violations are identified as harmful after manual inspection.