We investigate the internal representations that a recurrent neural network (RNN) uses while learning to recognize a regular formal language. Specially, we train a RNN on positive and negative examples from a regular language, and ask if there is a simple decoding function that maps states of this RNN to states of the minimal deterministic fnite automaton (MDFA) for the language. Our experiments show that such a decoding function indeed exists, and that it maps states of the RNN not to MDFA states, but to states of an abstraction obtained by clustering small sets of MDFA states into “superstates”. A framework for performing large scale systematic representation analysis between the two language models is discussed. Quantitative analysis surprisingly shows that linear decoding functions are suffcient for the task and an analysis of a range of abstraction functions is given. A qualitative analysis reveals new interpretations of how RNNs implement hierarchical priors during the language recognition task. Overall, the results suggest a strong structural relationship between internal representations used by RNNs and fnite automata, and explain the well-known ability of RNNs to recognize formal grammatical structure.