Non‐linear pricing, the fact that prices do not necessarily change in proportion to size, is a ubiquitous phenomenon. However, it has been neither particularly well understood nor well measured. Non‐linear pricing is of practical importance for statistical agencies who, in constructing price indexes, are often required to compare the relative price of a product‐variety of two different sizes. It is usually assumed that prices change one‐for‐one with package and pack size (e.g. a 1‐liter cola costs half as much as a 2‐liter bottle). We question the wisdom of such an assumption and outline a model to flexibly estimate the price‐size function. Applying our model to a large U.S. scanner dataset for carbonated beverages, at a disaggregated level, we find very significant discounts for larger‐sized products. This highlights the need to pursue methods such as those advocated in this paper.