Schools need evidence-based guidance on which measures in mathematics, administered under what particular set of conditions (e.g., time of year), provide the most useful prediction. The purpose of this study was to examine decision accuracy among commonly used screening measures with a priority toward identifying the least costly screening measures for predicting year-end mathematics failure. Predictors included existing demographic characteristics, the preceding year-end mathematics test score, and multiple measures administered during the study year including multiskill computation and concepts/applications measures, addition and subtraction for third grade, multiplication and division for fourth grade, and multidigit multiplication for fifth grade. Results supported the use of a single measure for screening. The preceding year's test score was superior or comparable in accuracy to current-year screening measures and was the lowest cost option (i.e., required no additional assessment time). Results cautioned against the use of multiskill computation and concepts/applications measures at all grade levels because of a high number of false-negative errors. The single-skill computation measures performed comparably to the preceding year-end test in overall accuracy. However, the single-skill probes outperformed all other measures in detecting students who would fail the year-end test, which is the most important function of a screening device. For most measures, the winter screening occasion offered the best predictive accuracy.