Abstract: A system of evaluation index for 100 m sprint race performance is constituted using fifteen indices including physique, function, constitution, psychology, technique, etc. Combining test indices of 100 youngster athletes aged 12 to 20 and experts' evaluation results on their 100 m sprint sports ability, the paper presents a neural network based sports ability evaluation model. The test indices were normalized and used as the neural network input, and comprehensive evaluation result of their 100 m sprint ability taken as the neural network output. Results of testing the trained neural network model were found to be satisfactory. The NN method overcomes insufficiencies of existing method and offers a new reference model in athelets selection.