Abstract: γ-ray spectral analysis serves as a core technology in nuclear radiation detection, environmental radioactivity monitoring, and nuclear security applications. However, in practical measurements, factors such as overlapping characteristic peaks, statistical fluctuations, and detector nonlinear response increase the difficulty and uncertainty of.manual spectral interpretation. To address these issues, this study proposes a γ-spectrum analysis method based on Fuzzy Neural Networks for nuclide identification and prediction of activity or enrichment, skipping traditional spectrum interpretation steps. First, fuzzy C-means clustering (FCM) is applied to partition spectral data, extracting membership degree features. These features are then fused with measurement condition characteristics and fed into a neural network to extract vector features, ultimately outputting radionuclide types along with predicted activity or enrichment. Experimental results demonstrate that the model achieves 100% accuracy in identifying ¹³⁷Cs, ⁶⁰Co and ¹³³Ba, and shows a mean value of absolute relative deviation of 2.02% of activity prediction. The RMSE for predicting the activity of ¹³⁷Cs, ⁶⁰Co, and ¹³³Ba respectively are 213.18 Bq, 944.31 Bq, and 10926.79 Bq. The relative deviations between their RMSE and the actual activity average values respectively are 2.89%, 2.11%, and 1.70%. The uranium enrichment prediction shows a mean value of absolute relative deviation of 11.84% and RMSE of 11.73%, indicating room for improvement. This approach effectively integrates fuzzy feature representation and measurement condition information, providing a viable solution for intelligent γ-spectrum analysis.