In this paper, we use the Chinese General Social Survey data to analyse the returns to upper secondary vocational education in China. To address possible endogeneity of vocational training due to omitted heterogeneity, we construct a novel instrumental variable using the proportion of tertiary education graduates relative to the entire population by year. Our main finding is that, although returns to vocational upper secondary education appear higher than returns to academic upper secondary education according to the Mincerian equation, the results from the instrumental variable method tell the opposite story: vocational upper secondary graduates face a wage penalty compared to academic upper secondary graduates. The wage penalty is confirmed by an alternative and more recent IV method - the Lewbel method (Lewbel, 2012). Our findings highlight the importance of properly accounting for endogeneity when estimating the returns to vocational education.