Rough set theory is a new mathematical tool to deal with vagueness and uncertainty. But original rough sets theory only generates deterministic rules and deals with data sets in which there is no noise. The variable precision rough set model (VPRSM) is presented to handle uncertain and noisy information. A method based on VPRSM is proposed to apply to fault diagnosis feature extraction and rules acquisition for industrial applications. An example for fault diagnosis of rotary machinery is given to show that the method is very effective.