I use an original dataset on the display inventories of several hundred eyewear retailers to study how firms' product-range choices depend on separation from rivals in geographically-differentiated markets. A two-stage estimation approach is used to model firms' initial location decisions and their subsequent choices of product variety. Per-firm variety varies non-monotonically with the degree of local competition. Holding fixed the total number of rivals in a market, a retailer stocks the widest variety when it is near a few other competitors. Firms with four or more rivals show substantially smaller product ranges. This suggests that business-stealing eventually dominates any clustering effects when there is intense competition in a neighbourhood.