In order to accurately predict the cost of construction project, according to the sample time dependent and nonlinear characteristics, this study has constructed a prediction model of construction project cost based on the geostatistics and support vector machine. Firstly, the cost data of construction engineering are collected, and then, the embedding dimension of time series is obtained by geostatistics according to time correlation of construction cost sample, and the construction cost learning samples of construction project, support vector machine is used to establish project cost prediction model, and test and analysis of the performance through the prediction example. The results show that the proposed model can effectively fit the construction cost of sample time correlation, and get higher accuracy of construction cost prediction, and the result is much better than the other models.