The invention relates to a filtering method based on a frequency distribution principle. The filtering method comprises the following steps: acquiring N groups of Raman spectrums, and calculating peak intensity storage arrays Ix of the Raman spectrums; dividing Ix into M sections; grouping the arrays Ix according to the sections, and counting the frequency of the sections; calculating a section K with the maximum frequency, and calculating a medium value Ixm (K) of the section; setting the maximum Raman spectrum intensity fluctuation deviation A, if the Raman spectrum intensity of the (N+1)th group is within a section [Ixm(K)-A, Ixm(K)+A,], judging that the Raman spectrums are effective, preprocessing, and storing in a register, or else judging that the Raman spectrums are ineffective, and abandoning, wherein the lengths of a plurality groups of Ix are fixed, the peak intensity of each group of newly acquired Raman spectrums is put into an array tail, one piece of data at the head of an original array is abandoned, N data is kept the latest, and overlapping is performed after L groups of effective Raman spectrums inside the register are achieved. The filtering method is remarkable in filtration effect and good in instantaneity, instantaneous noise interference can be effectively eliminated, and the interference resistance and the water temperature measurement precision of a Raman scattering ocean temperature laser remote sensing system can be improved.