This paper proposes a new radioactive nuclide identification method based on multi-layer perceptron to address the drawbacks of traditional methods such as multiple steps, slow speed, and low accuracy. The principal component analysis method is used to perform feature dimensionality reduction on the measured gamma spectrum dataset, and then a multi-layer perceptron model is established to quickly and accurately identify radioactive nuclides. The feasibility of the model for quantitative analysis of radioactive nuclides has been verified through experiments and compared with traditional methods for identifying radioactive nuclides. This method can avoid the occurrence of missed or misjudged types of nuclides, and does not require steps such as noise reduction, peak finding, and establishing a nuclide library, achieving fast and accurate identification of radioactive nuclides. Seven actual measured energy spectra composed of 137Cs, 60Co, and 133Ba radioactive nuclides were tested, and the results showed that the method achieved an accuracy rate of 99.91% for identifying radioactive nuclides.