The presence of reliability issues in nanometer semiconductor process technologies requires a paradigm shift toward robust and inherently fault tolerant digital signal processing systems. In this paper, we introduce a novel, low-complexity data-correction method to improve the system performance in presence of unreliable memories. In particular, we propose a novel statistical approach for best-effort error mitigation for data corrupted by defects in unreliable storage arrays. The correction function can be extracted from a statistical data-corruption model and a user-defined cost function, which must be chosen for the application at hand. To highlight the efficacy of our data-correction approach, we show an application example in the area of wireless communication systems.