The aim of this study is to gain insight into the strategies that might be necessary in a device for the automatic recognition of spoken sentences, through an experiment in which speech recognition is attempted by visual recognition of spectrograms by experienced experimenters. Spectrograms of a set of ten sentences, constructed from a vocabulary of 200 words, were prepared and the experimenters (the authors) attempted two tasks from visual examination of the spectrograms: (1) phonetic transcription of the sentences in terms of phonetic symbols or in terms of a partial feature specification; and (2) recognition of each sentence as a whole, using any available information, including the lexicon.