Aiming at the problem of longtime consuming and inconvenient maintenance of traditional rainfall measurement, based on the analysis of acoustic signal recognition technology, this paper proposes a rainfall measurement method based on sound recognition to simulate the nonlinearity of frequency domain demarcation and the mechanism of superimposing acoustic signals in the same frequency group in human ears. The Fourier transformed energy spectrum is passed through a Mel filter, and then the MelFrequency Cepstral Coefficients of the rain sounds are extracted as the eigenvector of rain sound signals. On this basis, a threelayer BP neural network is constructed, and the normalized sample data are used for neural network training. Finally, the test samples are used to identify the rainfall. Experimental results show that neural network can effectively identify the amount of rainfall on the basis of a small amount of sample training, which provides a theoretical basis for the application of acoustic signal recognition technology for more accurate rainfall measurement.