We report the determinations of glucose concentrations in human whole blood samples made using near-infrared Raman spectroscopy. Raman spectra of blood samples with above-physiological levels of glucose were acquired for 5 min through the wall of a cuvette via fiber optics. Partial least squares analysis was used to predict glucose concentrations in the samples. A root mean squared prediction error of 3.6 mM glucose was achieved with a correlation coefficient of 0.99 between reference and predicted values. This result is the first step in evaluating the potential of near-infrared Raman spectroscopy to perform blood glucose measurement with clinical accuracy. The technique is capable of measuring the concentration of other Raman-active blood constituents; as an example, bicarbonate was also measured. The method could eventually be useful for direct measurement of tissue analytes.