ABSTRACT The central problem addressed by this proposal is how the brain represents almost limitless objects to “catego- rize an unlabeled world” (Edelman, 1976). This problem is particularly acute in olfaction, because physical odor signals are not obviously dimensional. While instinctual meanings are imparted through the structure of neural circuits, learning mechanisms can theoretically select experience-relevant representations from circuits lacking a priori structure. By analogy, the vertebrate adaptive immune system defends against novel invaders by se- lectively amplifying B cells carrying strongly binding antibodies from a randomly produced pool. How unstruc- tured, combinatorial connectivity forms during development is not understood in any system. The numerical complexity of vertebrate olfactory cortex and other associative learning centers including the cerebellum, hip- pocampus, and entorhinal cortex complicates developmental studies. Instead, we propose to address this question in the numerically simplified olfactory learning center of arthropods, the mushroom body. While this brain area performs a function similar to the cortex or cerebellum, representing objects as combinations of sen- sory stimuli, the fruit fly mushroom body has only 2000 intrinsic neurons. These intrinsic Kenyon cells receive 3-10 olfactory inputs, with each cell receiving different inputs. The sets of inputs to individual Kenyon cells are unpredictable—knowing one input to a particular cell does not illuminate which other inputs that cell receives. To ask how unstructured connectivity pattern forms during development, we are first using a variety of tech- niques to manipulate the ratio between Kenyon cells and their incoming olfactory inputs in order to identify mechanisms that set the number of inputs to individual Kenyon cells. Second, we are testing whether variation in expression of cell surface molecules across pre- and post-synaptic cells allows variation in connectivity be- tween them. The quantitative complexity and density of sensory inputs to associative brain areas are theorized to “optimize” representations of sensory objects, with alterations in these patterns expected to degrade neural and behavioral distinctions among stimuli. If successful, we will produce a model of the developmental mecha- nisms that generate neurons responding to complex sensory stimuli that can be applied to studying develop- ment and organization of sensory inputs to vertebrate associative learning areas; this study will also create techniques to manipulate the wiring parameters of the Drosophila mushroom body to explicitly test how wiring parameters dictate stimulus representation and separation, and how their alteration alters perception and learning.