In personalized medicine, continuous biomarker values are often dichotomized to classify patients into target and non-target populations. Cast in the setting of normally distributed responses that are modeled linearly (such as diabetes and psychiatry), this article provides a method of inferring which thresholds correspond to target populations that benefit from the treatment. By providing simultaneous confidence intervals for efficacy corresponding to all candidate thresholds, our method allows for flexible decision-making, taking into consideration marketing potential based on both the size of the target population and efficacy in the target population. Under the assumption of the general linear model (GLM), advantages of our approach over the Jiang, Freidlin, Simon (2007) approach are that (1) formulation is clinically more meaningful, (2) imbalance in the data would not lead to misleading inference, (3) simultaneous confidence intervals are provided, (4) error rate and confidence level computations are precise, (5) simple SAS codes are available.