Various uncertainties of the FE modeling may have significant impact on structural identification and result in obvious discrepancy between FE model analysis and field test results. In order to eliminate epistemic uncertainty and aleatory uncertainty during model calibration procedure, which contributes to achieve reliable condition assessment and response prediction, two long-span bridges were taken for examples, and several critical issues were discussed. By means of field inspection and blueprint review, simultaneously combined with the precise modeling, uncertain parameter selection was carried out based on sensitivity analysis, and parameter optimization was eventually performed by the least square method to complete the working condition assessment and response prediction. Research in this paper consists of in-situ experiments, precise finite element modeling, model calibration, and model admissibility check. Discrepancies between FE model and field test results can be mitigated by efficient model calibration method, and initial design model generally shows 20% and 23% relative difference.