Spectrum auction is an important approach of spectrum distribution in cognitive radio networks. However, a single secondary user (SU) probably can’t afford the price of spectrum. Multiple SUs grouping together to participate in the auction as a whole is helpful to increase purchasing power. However, SUs could suffer from a new group cheating prob-lem, i.e., parts of users conspire to manipulate the auction by submitting untruthful bids. Existing auction mechanisms were mainly designed to be strategy-proof only for indi-vidual user and can’t deal with group cheat-ing. In this paper, a novel spectrum auction mechanism called COSTAG (COst Sharing based Truthful Auction with Group-buying) is proposed to address the group cheating prob-lem. COSTAG consists of a grouping rule to perform grouping and a payment rule to de-termine the market-clearing price in the spec-trum auction. Different from single-echelon pricing approach employed in existing works, a multi-echelon pricing strategy is designed to increase the transaction rate and optimize social profit for the auction. Comprehensive theoretical analysis shows that COSTAG can satisfy the crucial economic robustness prop-erties, both individual and group truthfulness. Simulations demonstrate that comparing with existing works, COSTAG can improve the system performance significantly.