Building ventilation is essential during COVID-19 to reduce airborne viral transmission risks, while accurately estimating air change rates remains quite challenging. Due to the tracer gas method's assumption of uniform air mixing in a space, which is often not the case. However, the existing mixing models such as mixing factor (K) and zone distribution effectiveness (Ez) are limited to decrease the uncertainty of this assumption since their reported data are subjective, rough estimates, and inconsistent across different standards (e.g., ASHRAE and AIHA). Therefore, in this study, a novel modified decay method (MDM) is developed, where a proposed uniformity index (Ui) is integrated into the original decay method (ODM) to quantify the non-uniform air mixing and thereby improve the estimation of air change rates. We validated the proposed method in a real classroom using CO2 as a tracer gas. Wherein, the spatial variations of CO2 concentrations were measured at various locations using an automated system of a mass spectrometer with a 16-position valve. Later, the estimated air change rates using both the ODM and MDM were compared with the measured ones. It was found that the proposed Ui significantly decreased the error caused by the uniform-mixing assumption of estimated air change rates using ODM from 25.6 % to 3.4 % estimated by MDM at the outlet location. Similarly, the average concentrations of 16 distributed sensors at the breathing level also showed a decrease in error from 25.0 % to 2.5 %. This study can be applied to improve ventilation performance in buildings.