It is evident that surface electromyography (sEMG)-based sensing approach for human–machine interfaces has some inherent limitations for applications requiring morphological changes information of deep-layered muscles, such as dexterous prosthetic hands. In this paper, the design, simulation, fabrication, and evaluation for a series of novel structured ultrasound transducers are conducted in order to develop a type of A-mode ultrasound transducers that overcome the drawbacks of the sEMG-based sensing. The transducers cover single-frequency and dual-frequency types. Their key parameters, the acoustic impedance and thickness of the matching layer, are simulated and verified by PZFlex. The parameters are designed as 0.3 times of the 1–3 composite piezoelectric’s acoustic impedance and 0.25 times of the wavelength, respectively. The characterizations of the dual-frequency transducers significantly outperform single-frequency transducers. The experiments of recognizing dexterous hand gesture are designed to detect morphological changes information of deep-layered muscles. The classification accuracy improvements with linear discrimination analysis are 7.3% and 4.7%, and with support vector machine are 14.1% and 13.4% for the horizontal stacked and annulus array. This preliminary study concludes that the dual-frequency transducers have huge potential for applications that need contraction information of deep-layered muscles over the single-frequency transducers, letting alone sEMG-based sensors.