Abstract: A BP neural network model and a radial basis (ＲBF) neural network model were set up respectively to predict the compressive strength of compacted earth blocks． The prediction model for the compressive strength of compacted earth blocks was based on 91 groups of experimental research data collected under the action of multiple factors by domestic and foreign scholars，and the influence of compaction pressure，moisture content，cement con-tent and height-thickness ratio on the compressive strength of compacted earth block was considered． The neural network predicted results were compared with the predicted results of regression analysis model and the experimental values． The results show that the prediction accuracy of artificial neural network model is better than that of regres-sion analysis model． And there is a complex nonlinear relationship that the regression analysis model cannot handle between the compressive strength of the compacted earth block and its mixture ratio，compaction pressure and height-to-thickness ratio． The overall prediction effect of BP neural network model is good，but it is easy to fall in-to local optimum． ＲBF neural network model can reliably predict the compressive strength of compacted earth blocks． The mean value of the ratio between the predicted results and the experimental values is 1. 007，with a standard deviation of 0. 085． The prediction model has a high accuracy and can effectively solve the complex non-linear relationship between compressive strength and its influencing factors，and it can provide reference for the de-sign of mix proportion for compacted earth blocks．

Key words: RBF neural network, BP neural network, compacted earth block, compressive strength, regression analysis model