Established in 2002, the Amazon Protected Areas Program (ARPA) supports 120 Conservation Units (CUs) in the Brazilian Amazon, covering 62 Mha. Here, we quantified the impact of ARPA support on reducing deforestation and CO2 emissions between 2008 and 2020. We started by examining critical methodological choices, often brushed over in the impact evaluation studies on protected areas (PAs). We then applied a covariate balancing method to control for variation in covariates so as to compare differences in deforestation between Strictly Protected (SP) and Sustainable Use (SU) CUs with and without ARPA support as well as to assess the influence of ARPA investment mechanism on the differential reductions. Next, we estimated total reductions in deforestation and CO2 emissions by using the Adjusted Odds Ratio. We found that ARPA support accounts for additional deforestation reductions of 9 % in SP CUs and 39 % in SU CUs in relation to non-supported CUs. The effects of ARPA investment mechanism were statistically significant for both categories of CUs. CUs plus Indigenous Lands (i.e., PAs) reduced by 21 % (2.0 ± 0.3 Mha) Amazon deforestation between 2008 and 2020. Of this total, ARPA CUs accounts for 264 ± 25 thousand ha, the equivalent of 104 ± 10 Mtons of CO2 emissions. If deforestation continues unabated, PAs will become the last citadels of the Amazon. However, protecting the Amazon only with PAs does not suffice. Additional investments in a comprehensive conservation policy mix are needed along with a monitoring and evaluation strategy to provide evidence on what works for effective and socially equitable forest conservation.