For non-invasively measurement of the components of human blood,dynamic spectrum method was used to measure hemoglobin concentration of volunteers for the first time. In-vivo measurements were carried out on 34 healthy volunteers,and their dynamic spectra were collected. To ensure the dynamic spectrum data to be valid,a number of experiments were carried out on the dynamic spectrum data. BP artificial neural network was used to establish the calibration model of subjects’ hemoglobin concentration values against dynamic spectrum data. Among 34 swatches,19 swatches were taken as calibration set and the other 15 as prediction set. For calibration set and prediction set,the correlation coefficient was 0.9831 and 0.9365 respectively. The biggest relative error of prediction is 7.5%, and the average relative error is 3.04%. The accuracy of measurement results can satisfy the demand for clinical application. Measurement results show that the influences of measuring conditions on spectra can be decreased effectively by dynamic spectrum method and this method can be applied to accurate non-invasive measurement of human hemoglobin concentration. It can be a good method for non-invasive blood analysis.