Modern fluorescence microscopes are capable of recording images with a reasonably resolved spectrum in each pixel, enabling advanced spectroscopic characterization of optically resolved structures. However, this requires access to adequate methods for resolving and discriminating spectral features or components of the generated hyperspectral images. There are several methods for accomplishing this, and they have become more and more sophisticated with the increase in computing power available in modern computers. One common way of doing spectral analysis is by linear spectral unmixing,1–3 which classifies the measured spectrum in every pixel in terms of a linear sum of input references. If the reference spectra are not known, a commonly used method is the principal component analysis,4 which separates the spectrum into its main (orthogonal) components, using statistics gathered from the image. Another approach that classifies the spectra is the spectral angle mapper algorithm.5–7 All of these methods have their drawbacks, e.g., they are difficult to use if the signals are weak with interfering background fluorescence or if there are large differences in relative quantum efficiency of different spectral components. The problems associated with the common signal processing tools prompted us to propose a spectral cross-correlation algorithm for hyperspectral images.8 The correlation coefficient8 is advantageous over, for instance, linear spectral unmixing,1–3 since it is statistically defined and scale invariant. It is limited to between and 1, with being , 0 and 1 . In spectral measurements, the measured intensities should never be negative, as intensity is dependent on the number of photons. As a result, the correlation coefficient should be larger than 0, assuming the background offset is reasonably small. Due to random noise in the measurement, the correlation coefficient will never be exactly 1. As will be shown, this method is well suited for the analysis of biological samples with a complex background as we can gain sensitivity by using all spectral ranges of the hyperspectral image, but we are at the same time able to discriminate the relative contribution from different spectral components as with the linear signal processing algorithms.