The pixel values would be different before and behind the geometric correction, which would lead to the different irradiance of dark pixel before and behind the geometric correction.Regarding the Lake Taihu as the study area and using the Landsat/TM, CBERS and HJ-1B satellite imageries, data collected at October 28, 2003, July 24, 2008 and April 25, 2009, respectively, as the data foundation, the paper researched and discussed the influences of geometric correction on the accuracy of dark pixel atmospheric correction algorithm and water leaving irradiance retrieval production.The results showed that the geometric correction would caused to 0.2, 0.47, 0.882 W/(m²·sr·μm) irradiance absolute deviations of dark pixel for Landsat/TM, CBERS and HJ-1B satellite imageries, respectively.There were 3.28%-9.81% relative biases for aerosol optical thickness estimated from the dark pixel algorithm.The influences of geometric correction would be relatively smaller on the Landsat/TM and CBERS imageries (relative bias mainly about 0 -6%), and relatively larger on the HJ-1B satellite imagery(relative bias mainly about 0-15%, and some regions up to 40%).As a result, it was well known that the atmospheric correction must be calculated before the geometric correction.