Local networks within the primary olfactory centers reformat odor representations from olfactory receptor neurons to second-order neurons. By studying the rules underlying mixture representation at the input to the antennal lobe (AL), the primary olfactory center of the insect brain, we recently found that mixture representation follows a strict elemental rule in honeybees: the more a component activates the AL when presented alone, the more it is represented in a mixture. We now studied mixture representation at the output of the AL by imaging a population of second-order neurons, which convey AL processed odor information to higher brain centers. We systematically measured odor-evoked activity in 22 identified glomeruli in response to four single odorants and all their possible binary, ternary and quaternary mixtures. By comparing input and output responses, we determined how the AL network reformats mixture representation and what advantage this confers for odor discrimination. We show that increased inhibition within the AL leads to more synthetic, less elemental, mixture representation at the output level than that at the input level. As a result, mixture representations become more separable in the olfactory space, thus allowing better differentiation among floral blends in nature.