Quantifying CO2 fluxes from lakes to the atmosphere is important for balancing regional and global-scale carbon budgets. CO2 emissions are estimated through statistical upscaling procedures that aggregate data from a large number of lakes. However, aggregation can bias flux estimates if the physical and chemical factors determining CO2 exchange between water and the atmosphere are not independent. We evaluated the magnitude of aggregation biases with moment expansions and pCO2 data from 5140 Swedish lakes. The direction of the aggregation bias depends on lake size;mean flux was overestimated by 4% for small lakes (0.01–0.1 km2) but underestimated by 13% for large lakes (100–1000 km2). Simple covariance-based correction factors were generated to adjust for upscaling biases. These correction factors represent an easily interpretable and implemented approach to improving the accuracy of regional and global estimates of lake CO2 emissions.