Nerf is a statistical tool for Named Entity Recognition (NER) based on the Conditional Random Fields (CRF) modelling method. The tool has been constructed as a part of the National Corpus of Polish project. It has been adapted to recognize tree-like structures of NEs (i.e., with recursivelly embeded NEs) using the Joined Label Tagging (JLT) method. The JLT method is a simple method of encoding NE structures as a sequence of labels. With this method various additional informations about NEs of categorical nature – type, subtype, type of derivation – can be encoded on the level of labels and subsequently recognized using the resultant CRF model. The tool can be configured to use various types of observations during the training and recognition process, for example: lexical informations from textual level, or grammatical informations from morphosyntactic level.