Gamma sources are routinely used to calibrate the energy scale and resolution of liquid scintillator detectors. However, non-scintillating material surrounding the source introduces energy losses, which may bias the determination of the centroid and width of the full absorption peak. In this paper, we present a general method to determine the true gamma centroid and width to a relative precision of 0.03% and 0.50%, respectively, using energy losses predicted by the Monte Carlo simulation. In particular, the accuracy of the assumed source geometry is readily obtained from the fit. The method performs well with experimental data in the Daya Bay detector.