ASURA is an experimental speech-to-speech translation system. It recognizes Japanese input speech, translates to English and German, and output synthesized voices. The main features concerning speech recognition, interface between speech recognition and language processing, and language translation are described. The performance evaluation of the system is reported, too. The result shows that the system has high recognition and translation abilities with fairly good efficiency. We had an international joint experiment on an Automatic Interpreting Telephone System connecting this system with other systems developed by Carnegie Mellon University and Siemens Corporation/Karlsruhe University. The result of the experiment is also reported.