Object-oriented classification of high spatial resolution imagery usually offers higher accuracy than traditional per-pixel approaches, as real-world features (objects) appear in those as groups of pixels with a relatively high internal variability – though generally lower than that of the whole image – due to the heterogeneous spectral response of their surface. Although a particular kind of object-oriented classification algorithm is currently implemented in GRASS GIS (the Sequential Maximum a Posteriori estimation or SMAP), it presents several drawbacks when compared to some proprietary alternatives. One of the most important shortcomings of SMAP is the fact that both the segmentation and classification phases have to be carried out on the same group of images: If textural information is used as an additional source of information, this usually increases the confusion between coarse-texture classes and border areas between different covers. Other interesting problem is the absence of any control over the size and shape of the objects created. While these problems might seem enough reason to turn attention to other pieces of software currently available in the market, other alternatives exist: One of such detour approaches (the one presented by this work) might be the use of existing vector layers as segments (objects) which may be classified using statistical techniques such as decision trees. This work presents an example of classification of historical aerial photographs into five broadly defined classes: (1) arable land, (2) pastures, (3) pastures with trees, (4) shrubland, (5) trees. Two different approaches are compared: (a) the standard use of the “i.smap” module, and (b) the classification of land cover polygons “borrowed” from the Spanish land parcel identification system (SIGPAC).