EvoSuite is a state-of-the-art search-based software testing tool for Java programs and many researchers have applied EvoSuite to achieve high test coverage. However, due to the high complexity of object-oriented programs, EvoSuite still may suffer several limitations in terms of test coverage achievement. In this paper, we conducted experiments to study the limitations of EvoSuite on the most recent SBST 2020 Tool Competition benchmark that consists of 70 classes selected from real-world Java projects. We have manually classified the uncovered branches of the target programs and reported corresponding limitations of EvoSuite. Finally, we discuss those limitations and how to resolve those issues to improve EvoSuite.