Robustness of neuronal activity patterns against variations in input intensity is critical for neuronal computations. We found that odor representations in the olfactory bulb were stabilized by interneurons that were densely coupled to the output neurons by electrical and GABAergic synapses. This interneuron network modulated responses of output neurons as a function of stimulus intensity in two ways: it globally boosted responses to weak odors, but attenuated responses to strong odors, and it increased the sensitivity of some output neurons, but decreased the sensitivity of others. These effects are closely related to strategies used in engineering to increase dynamic range. Together, they maintained not only the mean, but also the distribution, of activity across the population of output neurons within narrow limits, which is important for pattern classification. Neuronal circuits in the olfactory bulb therefore stabilize combinatorial sensory representations against variations in stimulus intensity by generic mechanisms.