In the era of big data, dynamic data have become more popular than static data because high volumes of data can be generated and collected at a rapid rate. Although rough set theory has been widely used as a framework to mine decision rules from information system, most of the existing algorithms were not designed to handle streaming data. Hence, in this paper, we present a system based on rough set theory to mine decision rules from streaming data. In particular, our rough set system processes data streams on two bases (namely, batch-based, and aggregated-based) with three models (namely, landmark, sliding window, and time-fading models) for a total of six combinations of stream processing and mining models (e.g., batch-based landmark model). Evaluation results on comparisons with existing works on several benchmark datasets show the benefits—in terms of both accuracy improvements and runtime reduction—and the practicality of our rough set system in mining data streams.