Computer transcription of Stenotype code offers the possibility of producing English text from speech via the Stenotype keyboard in real time. Reported are experiments directed toward designing a time-shared Stenotype transcription system that makes use of earlier work in Stenotype dictionaries and language processing. A content-addressing algorithm for direct-access storage, requiring a single access per retrieval, is presented. An existing experimental Stenotype dictionary program is used to implement on System/360 this algorithm and a dictionary-compaction technique. Transcription analysis indicates that the experimental design can reduce the average transcription error rate to six percent.