BACKGROUND Raman spectroscopy has advantages over infrared absorption spectroscopy. Combined with a novel multivariate technique, hybrid linear analysis (HLA), low prediction error is expected. METHODS A near-infrared (NIR) light source excited Raman signals, and a charge coupled device (CCD) camera was used to collect the signal. Samples were collected from 69 individuals for 7 weeks. The standard multivariate calibration technique, partial least squares (PLS) and HLA were both used to analyze the collected spectra. A Clarke error grid was used to evaluate the usefulness of the glucose measurement in serum. RESULTS The root mean squared error of prediction (RMSEP) for glucose in serum obtained with PLS is 21 mg/dL, and the RMSEP obtained with HLA is 17 mg/dL. In whole blood, the PLS RMSEP for glucose was 79 mg/dL, and HLA predictions had an RMSEP of 63 mg/dL. CONCLUSIONS The measurement technique was robust over the 7-week period. HLA was shown to generate a lower prediction error than PLS. The predictions by both PLS and HLA were clinically acceptable. The result with whole blood requires further improvement.