Generator inertia is an essential parameter in the analysis of frequency characteristics of power system and its online applications. The inertia of a generator can be identified in real time based on ambient active power and frequency signals measured while the generator is in routine operation. However,due to data quality defects of field measurements,the results of inertia identification are poor when applying existing algorithms to actual data. To solve this problem,the a priori variance of inertia identification results is defined based on spectral analysis and system identification theory. The a priori variance is calculated by three steps:reference system estimation,model parameter variance estimation and inertia variance estimation. The segments of ambient data are evaluated and selected before identification,which improves the accuracy of inertia identification. Data evaluation and selection results based on simulation data and field measurements verify the proposed method. The results show that the data segments with smaller a priori variance have higher accuracy of inertia identification.