This paper uses unique data on daily air pollution concentrations over the period 2001-2010 to test for manipulation in self-reported data by Chinese cities. First, we employ a discontinuity test to detect evidence consistent with data manipulation. Then, we propose a panel matching approach to identify the conditions under which irregularities may occur. We find that about 50% of cities reported dubious PM10 pollution levels that led to a discontinuity at the cut-off. Suspicious data reporting tends to occur on days when the anomaly is least detectable. Our findings indicate that the official daily air pollution data are not well behaved, which provides suggestive evidence of manipulation.