Testing to detect on-the-fly concurrency errors, such as atomicity violations caused by multiple shared variables, in a multi-threaded program is challenging because of the need to consider various factors, including correlations between variables and access event interleaving. An improved method for detecting atomicity violations during program execution is presented in this study. This method applies a straightforward approach to recognize a substantial number of related variables that make up atomicity as a single group and as one variable. This enhanced method was implemented, similarly to a test tool, and experimental comparisons were performed using a set of synthetic programs that modeled the execution of the seven representative multivariable atomicity violations. The results showed that compared with the original run of the test programs and a state-of-the-art detection method, the execution time increased by 1.07 and 1.05, respectively. This result included the accuracy of detecting all the multivariable atomicity violations undetected by the prior method.