We present a novel framework for set-valued data anonymization by partial suppression regardless of the amount of background knowledge the attacker possesses, and can be adapted to both space-time and quality-time trade-offs in a “pay-as-you-go” approach. While minimizing the number of item deletions, the framework attempts to either preserve the original data distribution or retain mineable useful association rules, which targets statistical analysis and association mining, two major data mining applications on set-valued data.