Price perception is extremely important for retailers. Customers assess the price of a product not only from the product’s own price history, but also from the prices of the product’s close variants. One particular kind of variant considered is the same product sold in different sizes, where a reduced unit price is generally expected for the ones sold in large quantities. Such price consistency between product variants could be important for customer experience, yet very challenging for retailers which carry millions of products with possibly missing and noisy catalog information. We propose a framework to measure pricing consistency between product size variants by retrieving product variants via search and extracting product size information with natural language processing methods. We evaluate three monotonic regression models that regularize the unit price instead of simple heuristics. To quantify the extent of price inconsistency, we define new metrics and demonstrate that one method can lower the inconsistency measure by up to 45% on the experiment sample set.