It is known that neurons can project topographically to their target area, and reciprocal projections back from the target area are typically aligned with the forward projection. However, the wide terminal arbors of individual axons limit the precision of such anatomical reciprocity. This leaves open the question of whether more precise reciprocal connectivity is obtainable through the adjustment of synaptic strengths. We have found that such a sharpening of projections can indeed result from a combination of biologically plausible mechanisms, namely Hebbian learning at synapses, continuous winner-take-all circuitry within areas, and homeostatic activity regulation within neurons. We show that this combination of mechanisms, which we refer to collectively as “sharp learning”, is capable of sharpening inter-area projections in a variety of network architectures. Sharp learning offers an explanation for how precise topographic and reciprocal connections can emerge, even in early development.