Protected areas (PAs) are supposedly key refuges for the world's remaining biodiversity. Our study site, Xishuangbanna, harbors a high proportion of China's biodiversity but is threatened by rapid deforestation and expansion of monoculture rubber. We quantified the success of Xishuangbanna's PAs in preventing deforestation. Most previous analyses of PA effectiveness have insufficiently accounted for biases arising from PA location and establishment, because they overlooked the importance of site‐matching in accounting for landscape change. We used matching methods to minimize such biases in comparing land use conversion rates inside and outside‐PAs. By 2010, Xishuangbanna had 3,455.5 km2 (~18%) designated as PAs. However, rubber occupied 22% of its land area and was expanding at a rate of 153.4 km2/year. Between 1988 and 2010, conventional analysis showed a deforestation rate of 9.3 km2/year. However, matching analysis showed a significantly higher rate of deforestation, 10.7 km2/year, which resulted in the deforestation of ~11% of PA's land. We argue that PAs were less effective than had previously been thought. The situation worsened from 2002 to 2010, when the deforestation rate within PAs was actually higher than that of outside PAs, although this difference was not significant. The designated higher levels of protection in ‘core’ zones were also unsuccessful in preventing deforestation. At current rates, within the next 50 years, a further 16% of PAs would be deforested in Xishuangbanna. This could even be an underestimate, as without intervention, drivers of deforestation tend to accelerate. Therefore, reviewing and strengthening current PA management policies is essential.