Attributes reduction is the main application of rough set theory. The present methods for reduction are mainly applicable to information systems with discrete values. For the continuous-valued attributes reduction, the common way is to get discrete intervals of values first and then transform the continuous values into the discrete ones. In such discretization, some information will be lost, which may influence the reduction. In this paper, a new approach for reduction of continuous-valued attributes (ReCA) is presented, which integrates the discretion and reduction using information entropy-based uncertainty measures and evolutionary computation. Experimental results show that the approach ReCA is effective for reduction of continuous-valued attributes, and can get less attributes and good precisions compared with the methods of rough set and C4.5 decision tree.