This paper uses computational experiments where bidders learn over nonlinear bidding strategies to compare outcomes for alternative pricing format for multi-unit multiple-bid auctions. Multi-unit multiple-bid auctions, in which bidders are allowed to submit multiple price-quantity bids, are promising mechanisms for the allocation of a range of resources. The main advantage of such auctions is to avoid the lumpy bid problem which arises when bidders can only compete on the basis of one bid. However, there is great uncertainty about the best auction formats when multi-unit auctions are used. The theory can only supply the expected structural properties of equilibrium strategies and the multiplicity of potential equilibria makes comparisons across auction formats difficult. Empirical studies and experiments have improved our knowledge of multi-unit auctions but they remain scarce and most experiments are restricted to two bidders and two units. Moreover, they demonstrate that bidders have limited rationality and learn through experience. This paper constructs an agent-based computational model of bidders to compare the performance of alternative procurement auction formats under circumstances where bidders submit continuous bid supply functions and learn over time to adjust their bids in order to improve their net incomes. The setting is for independent private values.We show that bidding behaviour displays more interesting patterns than is depicted in the theoretical literature and that biddingpatterns depend on the interplay between heterogeneity in the bidder population and the degree of rationing in the auction. Results indicate that the three auction formats have similar performance for most levels of competition but that their performances differ when competition is weak. This ranking is dependent on whether the population of bidders is homogenous or heterogeneous.