Research results presented in this dissertation are based on the texture as one of the four parameters (colour, texture, shape and motion) which can be used for the image description in the image retrieval systems, according to the new MPEG-7 standard. As the first step, texture features should be extracted to enable the use of texture for image retrieval purposes. Texture features extraction has been done by spatial-frequency analysis using Gabor filters. This analysis shows good performance for the texture features extraction, describing the image with one vector that contains 48 components. Spatial-frequency analysis was applied to Brodatz textures and image database was created. Brodatz textures are standard for the texture image processing. Software for the implementation of texture retrieval algorithm was produced. Algorithm is based on measurement of distance between feature vectors. Some examples of retrieval results are shown. Problems of the described method include the impossibility for the objective evaluation of algorithm performance and large number of retrieval results. To solve these problems, new texture retrieval algorithm based on the intersection of the results on the subimage level was realised. This approach has several advantages: image is divided in the non-overlapped regions (subimages), algorithm can be objectively evaluated, and the number of the results is suitably reduced. Image splitting algorithm and all the other algorithms are also realised by creating adequate software solutions. Retrieval results produced using new intersection approach confirm that this approach improves retrieval process. The method for texture features extraction was used in natural image analysis. Different tests of the retrieval algorithm accuracy have been performed, and the results are presented graphically comparing the total number of retrieved images with the number of accurate retrieval results. The retrieval accuracy for this case is acceptable. It confirms that texture features extraction method can be used for the analysis of the natural images. New approach, important for the MPEG-7 standardisation process, was introduced: classification of images according to their content. For this case, the retrieval of images is based on the comparison between the query image and available different image classes. Few image classes are created in a way that each class has the same image content (heads, faces, trains, flowers, snakes, portraits, and hills). For each image class, similarity between feature vectors is analysed, and one centroid per class is calculated. The accuracy of image classification algorithm was tested. Each image from each database was taken as a query image, compared with the 7 centroids, and classified according to minimum distance measure between the query feature vector and the centroid. The results show that the new algorithm for the image retrieval based on image content classification reduces retrieval process time and, in the same time, improves the retrieval accuracy.