We examine the effect of attending academically selective high schools on test scores, by leveraging administrative data that matches high school preferences of the population of urban middle school graduates in one Chinese prefecture in 2010 with high school student records. The standard admission channel is generally driven by merit subject to only nominal tuition fees, with contextual admission for disadvantaged students. An alternative admission channel admits lower-ability students subject to substantial selection-fees, retained by the under-funded schools. We combine a cumulative multiple-cutoff regression discontinuity design (RDD) with a within-cutoff normalizing-and-pooling fuzzy RDD strategy, based on publicly announced school-specific admission thresholds in the city-wide High School Entrance Exam (HSEE) scores. Multiple-cutoff RDD estimates show heterogeneous effects of attending schools with different degrees of selectivity, in a unified setting. Within-cutoff normalizing-and-pooling RDD allows admission thresholds to differ by willingness to pay the extra selection-fees and by eligibility for contextual admission. The estimated effects on high school leaving exam scores of attending elite schools vs normal public high schools, and of attending normal public high schools vs low-quality private high schools are insignificantly different from zero, for students who barely made it into the more selective school. However, the effect of attending the most selective flagship school vs elite schools, has a large negative and statistically significant effect, which is more pronounced for girls, for students from the semi-urban area according to hukou (household) registration, and for students who performed relatively badly in the science track subjects in the HSEE.