The heavy metal content in human hair can reflect the health status of a human body. A method for using hyperspectral data to detect the heavy metal element chromium content in human hair is put forward. The processing such as continuum removing and absorption characteristic parametrization is performed on the transmittance wavelength curves of some hair samples. Taking the chromium content detected by a chemical method as the standard data, the accuracy of chemical detection can reach above 90%. Then, the BP artificial neural network is trained and the model is optimized by adjusting the network number of hidden layer nodes, the number of nodes and the activated function. The experimental calculation results show that the artificial network which has one hidden layer and seven or nine nodes has a better prediction result. Both internal precision and external precision of the artificial network are evaluated by using the statistical experimental results. The Cr metal in human hair is sensitive in the wavebands from 1380 nm to 1550 nm, from 1880 nm to 2100 nm and from 2120 nm to 2210 nm. The trained BP artificial neural network has a prediction error RMS of 13% and an accuracy of 87%. The experimental results show that the hyperspectral technology can be used to measure the chromium content in human hair nondestructively and quickly.