This paper describes a vision-based flexible vibratory feeding system (VBFVFS) for agile manufacturing, which mainly consists of a decupled vibratory feeder, a machine vision system, displacement sensors, power amplifiers and a computer system. Its major feeding parameters such as the vibration angle, frequency, acceleration amplitude, and phase difference all can be easily adjusted online by software means. Consequently, the system not only can feed a wide range of parts without any retooling of the feeder, but also can feed various parts made of different material in a way close to the optimal state. The VBFVFS also has intelligence. It can find out the natural frequencies of the decoupled vibratory feeder by means of automatic frequency response analysis and hence determine the best frequency for the driving signals. It can also find out any parts jamming and eliminate it by making the parts move in reverse directions. Prototypes of the decoupled vibratory feeder have been successfully developed. Experimental investigation is carried out based on the prototypes and the results agree well with the simulation results. Moreover low-cost vision technique is also discussed for the commercialization of the vision-based flexible vibratory feeding system.