<p>Most of the real-world data suffer from uncertainty, imprecision, or inconsistency. Ignoring this fact may lead to biased information/knowledge extracted from such data. In this paper, we are interested in mining frequent patterns from imperfect data where uncertainty is modeled through the Dempster–Shafer theory. To this end, we propose a slight modification to the definition of the support measure to consider the tuple-level uncertainty. Then, a new algorithm called DP-FIMED is introduced. It is based on a two-level pruning technique that speeds up the generation of frequent itemsets. Intensive experiments were performed on real and synthetic evidential databases. The performance study showed interesting results compared to existing methods.</p>