Finding useful texture based features is often hard, even if there are clear patterns in the material at hand, but it will be a very useful contribution to any classification process as the texture might be uncorrelated to other more easily found features. So even if the classification rates can not compete with the ones accomplished with other approaches, they might give valuable new information.This thesis looks at the use of the Local Binary Pattern (LBP) texture analysis tech- nique in Prognostic Classification of Ovarian Cancer based on microscopy images of cell nuclei. The textures in this data set do not contain information that makes it possible to classify by human visual inspection. The approach used is not specific to the present data set, but could also be used on most other texture classification problems.Differences between LBP and Gray Level Cooccurrence Matrixes (GLCM) are de- scribed, feature selection in general is discussed, and pseudocode for some of the feature selection algorithms is presented.Extensions to regular LBP are proposed and tested together with the regular version. Grouping of the LBP data based on the nuclei sizes and gray levels and dividing into new features is tested. Other possible approaches are also outlined.The resulting classification rates are low. The reason for this is discussed and the main theory is that too much data is included in each feature. A version of regular LBP that only looks at dark areas is tested and shows some promising results.LBP looks like a promising tool for the future, but this thesis should give a good indication of what problems to be aware of. Any future success using LBP on this data set will probably depend on finding the areas whitin nuclei which have patterns that differ between the prognostic groups, as there does not seem to be any pattern present throughout most of the nuclei.