This article reported researches on the pattern recognition of finger key-press gestures based on surface electromyographic (SEMG) signals. All the gestures were defined referring to the PC standard keyboard, and totally 16 sorts of key-press gestures relating to the right hand were defined. The SEMG signals were collected from the forearm of the subjects by 4 sensors. And two kinds of pattern recognition experiments were designed and implemented for exploring the feasibility and repeatability of the key-press gesture recognition based on SEMG signals. The results from 6 subjects showed, by using the same-day templates, that the average classification rates of 16 defined key-press gestures reached above 75.8%. Moreover, when the training samples added up to 5 days, the recognition accuracies approached those obtained with the same-day templates. The experimental results confirm the feasibility and repeatability of SEMG-based key-press gestures classification, which is meaningful for the implementation of myoelectric control-based virtual keyboard interaction.