This article develops a simple model of teacher value-added to show how efficient use of information across subjects can improve the predictive ability of value-added models. Using matched student-teacher data from North Carolina, we show that the optimal use of math and reading scores improves the fit of prediction models of overall future teacher value-added by up to a third for reading and a tenth for a composite measure (math and reading combined). Efficiency gains are greatest when value-added must be calculated on only 1 or 2 years of data. The methods employed are flexible and can be expanded to incorporate information from other subject or subitem test metrics. (Contains 4 tables and 4 notes.)