Abstract: Traditional noise data processing methods have certain requirements for input data, and have a long running time. However, there are some problems of manual adjustment noise data of coal mine gas sensor such as less quantity, poor quality, inconsistent time and easy to be affected by environment, it is difficult to filter the noise data by traditional noise processing methods. For the above problems, a manual adjustment noise data processing method for coal mine gas sensor is proposed. The data average value is used to fill the missing value of concentration data of coal mine gas sensor; the feature set and sample set of concentration data of coal mine gas sensor are constructed by using multi time granularity; five curve fitting functions, namely Gaussian function, mixed Gaussian function, binomial function, trinomial function and piecewise binomial function, are used to fit manual adjustment noise data, and parameters of the fitting function are determined based on the least square method, and the optimal fitting function is obtained according to fitting effect; through analysis of manual adjustment noise data, it is concluded that the noise data is related to the slope, peak and difference of gas concentration before and after adjustment, according to these basic characteristics, the manual adjustment noise data is identified and deleted. The experimental results verify effectiveness of the method.