There are three forms of mercury in water: elemental mercury, inorganic mercury and organic mercury. Methylmercury is the main organic mercury form, and is much higher than that of elemental mercury and inorganic mercury. Cold vapor atomic fluorescence spectrometry(CVAFS) is the recommended method for measuring methylmercury in the water. CVAFS is an element analysis method developed from atomic emission and absorption spectrometry. After years of development and improvement, it has become one of the most commonly used technologies for element analysis. It is widely used in environmental protection, life science, geology and other fields with the characteristics of high sensitivity and low detection limit. However, affected by the background noise of the excitation light source, electronic components of the detection instrument and the separation effect of the chromatographic column, the signal of CVAFS will haveproblems such as baseline drift and signal tailing, which will seriously influence the peak area calculation of the CVAFS’s data and the quantitative analysis of trace methylmercury. Baseline drift is the most critical problem. At present, improving analogue device parameters and digital baseline estimation are two important ways to solve baseline drift. In terms of improving the parameters of the analogue device, there are hollow cathode mercury lamps and closed-loop controlled hot cathode low-pressure mercury lamps with disadvantages such as complex experimental equipment and high cost. The digital baseline estimation includesthe least square method, difference fitting method and so on, as all of them have weaknesses like unstable baseline estimation and inaccurate content calculation. Thus, a digital baseline estimation method based on wavelet transform was proposed. Firstly, by analyzing the microscopic signal of CVAFS and baseline drift of methylmercury, the mathematical model of the signal of CVAFS and baseline drift wasestablished. Secondly, according to the characteristics of the signal of the CVAFS model and wavelet transform, an appropriate mother wavelet model was established. The mother wavelet model was convoluted with the baseline drift model, and the convolution result was always zero. Theoretically, it indicated that the baseline drift wouldbe eliminated after wavelet transform. Thirdly, taking 100 pg standard sample methylmercury as an example, the experiments verified that wavelet transformation could eliminate baseline drift and solve the problem of signal tailing. Finally, under the condition that the relative standard deviation (RSD) of the instrument is 1.29%~3.40%, the experiments were repeated for 5 times for standard methylmercury solutions of 0, 10, 20, 50, 100, 500 and 1 000 pg, and the calibration curves of the average peak area before and after wavelet transform were established respectively. The calibration curve’s correlation coefficient (R²) is increased from 0.994 to 0.997 after the wavelet transform. The experimental results showed that this method could effectively eliminate the influence of baseline drift and signal tailing and improve the system’s measurement accuracy.