We present a new integrated approach for natural language call routing based on stochastic language models. The system learns automatically from examples to direct a call to the appropriate destination within a call center. It employs stochastic language models for each call destination. The language models are generated by a language model adaptation algorithmbased on the minimum discrimination information. The call routing approach is consistent with the stochastic pattern recognition paradigm and can be easily integrated in a automatic speech recognition (ASR) system. It results in over 92% correctly routed calls in a natural language call center application.