Abstract: The spectral parameters of the Cole-Cole model can improve the resolution of comprehensive interpretation of time-domain induced polarization (IP) data, contributing somewhat to the exploration of metal deposits. Applying the backpropagation neural network (BPNN) model to the prediction and inversion of spectral parameters can avoid high computational complexity to improve the inversion speed. Moreover, the BPNN model can fully explore the utilization efficiency of time-domain IP data to enrich the characteristic information of subsurface ore bodies. Based on this, this study derived the mathematical expression of the time-domain apparent polarizability attenuation curve using the digital filtering algorithm. With the mathematical expression as the forward/inverse model, this study comparatively analyzed the impacts of four factors-the sample size of the training set, the number of neurons in the input layer, the node number of hidden layers, and the number of hidden layers-on the training and inversion effects of the BPNN model, determining the optimal model. Furthermore, this study trained the BPNN model using time-domain IP data from eight time windows. Finally, this study applied the trained BPNN model for prediction and inversion based on the measured time-domain IP data. The results indicate that the BPNN model is feasible in inverting spectral parameters based on both theoretical and measured datasets, manifesting high inversion accuracy and minor errors. Overall, the results of this study can assist in distinguishing paragenetic and associated minerals and reducing misinterpretation.