Norway spruce [Picea abies (L.) Karst.] heartwood and sapwood have differing wood properties, but are similar in appearance. An investigation was made to see whether near-infrared spectroscopy (NIRS) could be used with multivariate statistics for separation between heartwood and sapwood in dry state on tangential longitudinal surfaces. For classification of wood into sapwood and heartwood, partial least square (PLS) regression was used. Orthogonal signal correction (OSC) filtering was used on the spectra. This study shows that a separation of sapwood and heartwood of spruce is possible with NIR spectra measured in a laboratory environment. The visible-wavelength spectra have significant influence on the predictive power of separation models between sapwood and heartwood of spruce. All 44 specimens in the calibration set were correctly classified into heartwood and sapwood. Validation of the model was done with a prediction set of 16 specimens, of which one was classified incorrectly.