Dissolved gas analysis (DGA) is an important method for on-line monitoring of power transformer. The traditional calibration method used in determining the calibration curves of gas dissolved in transformer oil is the segmented polyline method,but its smoothness, repeatability and overall fitting accuracy are not satisfactory. In order to solve these problems effectively, a novel calibration method for the gas dissolved in transformer oil based on Levenberg - Marquardt algorithm is proposed in this paper. Through establishing the gas dissolved in oil monitoring system, the data from five kinds of oil gases: H2, CH4, C2H6, C2H4 and C2H2 are collected and the proposed algorithm is applied effectively in curve fitting of calibration curves on these five kinds of gases. The experimental results show that this method can overcome the deficiencies of the traditional method and the fitting accuracy meets the requirements of State Grid’s standards very well.