In this paper, a novel multi-level adaptive lines of communication extraction method for multispectral images is presented. The method takes into account both spectral and spatial characteristics of the data on different levels of processing. The principal background classes are obtained first using K-means clustering. Each pixel is examined next for classification using a minimum distance classifier with principal class signatures obtained in the previous level. In the next level, the neighborhood of each unclassified pixel is analyzed for inclusion of candidate classes for use as endmembers in a spectral unmixing model. If the list of candidate background classes is empty, the conditions for their inclusion are relaxed. The fractions of backgrounds and lines of communication signatures for the unclassified pixels are determined by means of linear least-squares method. If the results of unmixing are not satisfactory, the candidate clusters list is renewed, and unmixing is repeated. The lines of communication detection within each pixel is performed next. The line segments detection parameters are initialized, directional confidence is calculated, and line segment tracking is initialized. The line segments are incremented until the composite confidence becomes too low. At the end, segment connection, and lines of communications identification is performed. The proposed method was successfully applied to both synthetic and AVIRIS hyperspectral data sets.