Odors are important cues for identification of many types of objects that animals need for survival. Natural odors are typically mixtures of up to a few dozen chemical components. Important information about odor 'objects' is often encoded in the ratio of components in the mixture. However, this odor mixture problem is complicated by two factors. First, many times the information channel is composed of a submixture of the overall mixture composition. Second, the ratios of components in the submixture can vary from one object to the next, which means that animals must learn to 'generalize' across a range of variation among objects that mean the same thing. Floral odors, for example, are important for honey bees to locate nectar and pollen sources that their colony needs for survival. Honey bees need to learn about the range of variation in odor composition so that they can optimally include flowers that have resources and exclude flowers with similar odors but which do not have nectar or pollen. I argue that nonassociative and associative plasticity in neural networks involved in early sensory coding is critical for extracting the relevant features of an odor mixture and setting up categories of odor objects. This plasticity can enhance decisions about pattern matching and multimodal associations in later processing in the brain.