Database marketers are often tasked with identifying high-value customers, whose profit exceed a predetermined value (PV) for an effective solicitation. The standard method for building a High-value Customer Model first predicts the profit variable using the ordinary regression model. Then, individuals whose predicted profit is greater than PV are selected for the solicitation. The purpose of this article is to introduce a better method for targeting high-value customers - in terms of smaller prediction error. The method uses a binary version of the profit target variable vis-a-vis PV, and models the binary-profit variable with logistic regression analysis. Then, individuals whose estimates of likelihood of profit greater than PV are selected in descending order for the solicitation. I discuss a real case study to compare and contrast the current and new methods to show that the latter produces smaller prediction bias - when the sample size is large and the model is misspecified.