We demonstrate the use of Raman spectroscopy to measure the concentration of many important constituents (analytes) in serum and whole blood samples at physiological concentration in vitro across a multipatient data set. A near-infrared (830-nm) diode laser generates Raman spectra that contain superpositions of Raman signals from different analytes. Calibrations for glucose, cholesterol, urea, and other analytes are developed by use of partial least-squares cross validation. We predict six analytes in serum with significant accuracy in a 66-patient data set, using 60-s spectra. The calibrations are shown to be fairly robust against system drift over the span of seven weeks. In whole blood, a preliminary analysis yields accurate predictions of some of the same analytes and also hematocrit. The results hold promise for potential medical applications.