The Japanese Earth Resources Satellite (JERS-1) Synthetic Aperture Radar (SAR) L-band HH polarization data over northern Australia were acquired in 1996 as part of the Global Rainforest Mapping (GRFM) experiment by the National Space Development Agency of Japan (NASDA). The data were mosaiced by NASDA and re-sampled to a resolution of 100 m×100 m per pixel. As the mosaiced data is a preliminary product, the problem of calibration, including the removal of effects of look angle variations within each scene and the relative calibration among multiple scenes due to different acquisition dates, has not been dealt with sufficiently, which greatly degrades the quality of the product. It is understood that precise calibration in mosaiced data is of necessity in order to retrieve accurate geophysical parameters. Nevertheless, the preliminary mosaiced data is used in the paper to investigate the potential of using SAR data for vegetation mapping in a regional or global scale. Pixel-based mapping using SAR data has shown limited success due to speckle. However, area analysis techniques allow the computation of reliable statistical and textural measures, thus providing a base for pursuing radar classification. In this paper, an image is first segmented using a Gaussian Markov random field model, so that area analysis can be pursued. An unsupervised classification method is developed in which as many as 12 statistical and textural measures are used to group segments with similar textural measures into the same class in a fashion of maximum likelihood. For a pre-given number of 50 as the final number of classes in the unsupervised classification iteration, the average B- distance among these 50 classes is 1.96 when all 12 textural measures are used to compute the distance.