This is one of the first studies to investigate gender achievement gap in the National College Entrance Exam in a typical municipality in China, which is the crucial examination for the transition from high school to higher education in that country. Using ordinary least square model and quantile regression model, the study consistently finds that the gender difference on average is not significant in mathematics, but is significantly negative (females worse) at the top of the distribution and is significantly positive (females better) in Chinese and English. Negative gender gaps are found among some disadvantaged subgroup such as urban female students with sibling. For students at the bottom of the distribution, boys perform significantly worse than girls.