Land type classification using SAR data is an area of current interest and research. In the paper, a quantitative analysis is made for JERS-1 SAR imagery, and a new classification technique is applied to determination of land types in the SAR images. The authors utilized an unsupervised neural network to provide automatic classification, and employed an iterative algorithm to improve the performance. First, S/N and statistical properties are evaluated for each land type, and it is shown that the images have enough quality for classification. Then, learning vector quantization (LVQ) is applied to unsupervised classification of SAR images, and the results are compared with those of the migrating means method. Results show that LVQ outperforms the migrating means method which classified most of the pixels into two classes. To further improve the performance, an iterative algorithm, where the SAR image is reclassified using the maximum likelihood (ML) classifier, is applied. It is experimentally shown that this algorithm converges, and significantly improves the performance of the unsupervised LVQ method while preserving the advantages of automatic operation inherent in unsupervised techniques.<<ETX>>