Several successful approaches to speech recognition have been proposed. Most of them involve time alignment which requires substantial computation and considerable memory storage. In this paper, we present a neuro-fuzzy approach to speech recognition without time alignment. This approach is a powerful method for selecting reference templates; therefore, considerable memory storage is alleviated. In addition, it greatly reduces substantial computation in the matching process because it obviates time alignment. Base on this approach, a Mandarin speech recognition system without time alignment is implemented. Two databases were utilized for verifying its performance. An encouraging experimental result confirms the effectiveness of the proposed neuro-fuzzy approach to speech recognition without time alignment.