The Kalman filter is increasingly applied to estimate air change rates. However, the filter’s reliability in this application requires further validation, as research on its computational accuracy remains limited. This study aims to provide experimental evidence for the computational accuracy of the Kalman filter. To this end, a multi-metric evaluation system based on estimation error and convergence time was developed. Using this system, the filter’s computational accuracy was evaluated by comparing its air change rate estimations against benchmark values across various experimental conditions. The results highlight the significant influence of indoor CO2 emission rate, ventilation condition, and initial indoor-outdoor CO2 concentration difference on the filter’s computational accuracy. Furthermore, the results demonstrate that the computational accuracy ultimately depends on the indoor-outdoor CO2 concentration difference and the uniformity of indoor CO2 distribution. Based on an estimation error threshold of 15% and a convergence time threshold of 50 minutes, this study identifies the following scenarios as unsuitable for applying the Kalman filter to air change rate estimation: 1) when the initial indoor-outdoor CO2 concentration difference is below 16 ± 12 ppm; 2) when an indoor CO2 source is present but its emission rate is low (below 0.0150 m3/h) coupled with an air change rate above 3.52 ± 0.07 h-1; and 3) when no indoor CO2 source exists and ventilation is strong (air change rate above 3.52 ± 0.07 h-1). This study defines the effective application scope of the Kalman filter, providing a critical reference for its appropriate use in building ventilation estimation.