This study shows selected statistical learning models for prediction VO 2 max based on 20 m shuttle run and selected somatic parameters. The research were conducted on 80 male physical education students (age: 21.4±1.6). Predictive models were determined from 65 patterns and tested using 15 patterns. The 9 variables were used to build the models, including one dependent variable and 8 independent variables. The prediction methods used were as follows: ordinary least square regression, Ridge regression, Lasso regression, multilayer perceptron and network with radial basis functions. The research show that the model with the least error of prediction is the network with radial basis functions generating the RMSE CV = 4.04 error. The best method for new data generate an error on the level of 3.04 compared to the real values.