With the adoption of Learning Management Systems (LMSs) in educational institutions, a lot of data has become available describing students’ online behavior. Many researchers have used these data to predict student performance. This has led to a rather diverse set of findings, possibly related to the diversity in courses and predictor variables extracted from the LMS, which makes it hard to draw general conclusions about the mechanisms underlying student performance. We first provide an overview of the theoretical arguments used in learning analytics research and the typical predictors that have been used in recent studies. We then analyze 17 blended courses with 4,989 students in a single institution using Moodle LMS, in which we predict student performance from LMS predictor variables as used in the literature and from in-between assessment grades, using both multi-level and standard regressions. Our analyses show that the results of predictive modeling, notwithstanding the fact that they are collected within a single institution, strongly vary across courses. Thus, the portability of the prediction models across courses is low. In addition, we show that for the purpose of early intervention or when in-between assessment grades are taken into account, LMS data are of little (additional) value. We outline the implications of our findings and emphasize the need to include more specific theoretical argumentation and additional data sources other than just the LMS data.