High resolution remote sensing imagery offers one possible means to discriminate the constituent land cover components within savanna woodlands, with differing economic potential and ability to sequester carbon. Recent work on tropical savannas in Belize suggests that per-feature classification of high-resolution satellite imagery such as IKONOS produces more accurate classifications of vegetation than conventional per-pixel classifiers, although low density woodlands have still proven difficult to classify. This study focused on the pine woodlands of those areas, seeking to characterise them by properties such as their constituent tree density and canopy cover. A bottom-up, rule-based classification algorithm was developed within a training subset of the IKONOS imagery for identifying and delineating individual tree crowns. Fuzzy classification rules with a distance element were used for capturing the spatial distribution of the identified trees and for delineating extended pine woodlands. The classification algorithm was applied on a validation subset to examine whether it can be automated. The accuracy for both, the training and validation subsets was assessed using a visual interpretation that has been created with image interpretation skills. The delineated pine woodland boundaries were also compared visually to classifications created by previous workers within the same study area. The results for identifying individual tree crowns were satisfactory while their delineation was disappointing. The delineation of extended pine woodlands was adequate, however gap creation was observed within the validation subset. Visual comparison to previous work shows that the bottom-up approach provides better delineation of pine woodlands. Even though this classification algorithm may be appropriate for characterizing low density savanna woodlands different approaches and methods should be tested as well, mainly for delineating individual tree crowns.