This study explored the role of elementary school students’ computational thinking skills and how they were related to the performance in a self-designed robotic game <STEM Port>. This game required the operation of robots on the large world map in which the players carried out competitive tasks with block coding. Five computational thinking dimensions in terms of problem-solving skills were examined which includes Algorithm, Evaluation, Decomposition, Abstraction, and Generalization. Among total of 99 six-graders, there were 66 and 33 students with high and low algorithm skill respectively. The regression results showed that students with high algorithm skill could have overall better performance in the game than those with lower algorithm skill, especially in decomposition and generalization dimensions in the four rounds of the game. It was also found that the high algorithm students’ performances in the beginning and end round were predicted by their perceptions of computational thinking skills, however, none of low algorithm students’ perceptions of computational thinking skills were associated with their performances. It implied the importance of CT education to the students which would lead them to have better performance in the complex problem-solving situations such as the fast advancing world lies before them.