In order to quickly obtain and evaluate the comprehensive performance of concrete, a prediction model of the comprehensive performance of concrete was established based on the relevance vector machine (RVM), in which six main factors affecting the comprehensive performance of concrete were selected as input data and the comprehensive performances (28 d strength, slump extension and apparent density) of concrete were selected as the output data. Then the model was used to predict 5 groups of predicting samples through fitting traning of 14 groups of learning samples. The results show that under the same sample conditions, compared with BP neural network model, RVM model has higher prediction accuracy and less discreteness. Compared with the actual value, the average relative error of concrete comprehensive performance index predicted by RVM model is obviously smaller than that predicted by BP neural network model, which further verifies the reliability of RVM model to predict the comprehensive performance of concrete, and has good promotion value.