Recent advances in information technologies create numerous opportunities for retailers to turn customer information into additional pro ts by targeted pricing: charging different prices to di erent market segments based on customer demographic variables. However, rigorous theoretical analysis regarding what is the pro t-maximizing set of variables remains scarce. This study develops a game-theoretic model to investigate this question. Results of this model suggest that a monopoly seller should use pricing variables with high explanatory power of its demand: i.e., variables with large demand coe cient or high variance. In the duopoly case, this model suggests that the value of price discrimination is the same as that in the monopoly case when only one rm uses that pricing variable. When two symmetric rms simultaneously use a pricing variable, the value of price discrimination may be higher or lower. When two rms sell substitutes and the demand is a ected by that variable in the same direction, VOPD is higher. In contrast, when the demand is a ected by that variable in the opposite direction, VOPD is lower. When rms sell complements, these e ects are reversed. This model is applied to horizontal di erentiation and vertical di erentiation to explain the puzzle that in practice similar products are often priced based on di erent variables.