The problem of high school dropouts has been studied for decades, but utilizing readily obtainable student data and data mining can aid school leaders to more accurately detect which students will likely drop out. This early warning information can be used by educators as early as sixth grade to help identify potential high school dropouts and students who will not graduate on time and intervene more efficiently and effectively with those students. The purpose of this nonexperimental correlational study was to use longitudinal data from a mid-sized school district from two cohorts to support the creation of a dropout early warning system to predict both sixth-grade and ninth-grade students who are at risk for not graduating on time. The statistical models utilized to identify the most accurate indicators were logistics regression, linear discriminate analysis, and quadratic discriminate analysis. The variables identified in the ninth-grade models as able to predict students who would not complete high school within four years were: if a student did not receive enough credits to advance to the tenth grade, did not attend school at least 90% of the time, was suspended from school, had multiple school moves in the ninth grade, and gender. The sixth-grade variables identified as able to predict students who would not complete high school within four years were: if the student was suspended from school, had multiple school moves, did not pass English, and gender. The model identified as having the lowest false-positive rate and a relatively high accuracy for ninth grade was the downsampled QDA model with the lowest false-positive rate of 0.43 and an accuracy of 75%. For sixth grade, the downsampled LDA model had the lowest false-positive rate at 0.34 and an accuracy of 67%. Keyword 1: dropout Keyword 2: late graduate Keyword 3: predict dropouts Keyword 4: early warning