In order to obtain reliable investment estimates in the early stage of the metroconstruction project for project decision-making and project cost control， on the basis of an-aly zing andsummarizing the relevantcharacteristics of the civilengineering structure ofmetro tunnel， this paper uses the powerful function mining ability of gene expression pro-gramming to find out the non-linear relationship between the uncertain factors affecting thecivil engineering cost of metro tunnel and the cost of extension meter.The prediction modelis constructed to estimate the civil construction cost of station tunnel， and the civilcon-struction cost of the whole metroline tunnel is estimated . 16 groups of typical engineeringdata were selected as training samples and verification samples.After setting the compo-nents of GEP algorithm， an algorithm programwas written byMatlab2016b to solve theprediction model.After repeated calculations， a good fitting function with fitness value of 1214.04 was found.The example results show that the relative error between the predictionmodel and the actual value is 7.78%， which meets the requirement of prediction accuracy10%.In addition， the comparison between GEP prediction model and ANN model showsthat the verified sample prediction results of GEP prediction model have better predictionaccuracy.