Constantly i ncreasing maintenance costs impose optimal maintenance policy planning. One possible way which helps to minimize maintenance costs and prevent bus fleet availability is analysis of historical maintenance records, which contain information about failures and performed repairs. In many cases this data have free text form and their analysis require individual log-by-log examination of their content. In order to automate this process, text mining methods can be applied. But, accuracy of the analysis depends on data quality and employed methods and should be tested before using this approach. This is especially important when the service decisions, which influence safety and maintenance costs, are made on this basis. The aim of this paper is to determine whether existing and currently used text -mining methods are sufficiently accurate to be used in classification of unstructured urban bus maintenance and repair data. For that purpose the case study and literature review has been conducted. The study shows great capabilities of proposed classification model. The model has 99% of accuracy and can be applied to support maintenance decisions.