This study compared the effects of three different feedback formats provided to sixth grade mathematics students within a web-based online learning platform, ASSISTments. A sample of 196 students were randomly assigned to one of three conditions: (1) text-based feedback; (2) image-based feedback; and (3) correctness only feedback. Regardless of condition, students solved a set of problems pertaining to the division of fractions by fractions. This mathematics content was representative of challenging sixth grade mathematics Common Core State Standard (6.NS.A.1). Students randomly assigned to receive text-based feedback (Condition A) or image-based feedback (Condition B) outperformed those randomly assigned to the correctness only group (Condition C). However, these differences were not statistically significant (F(2,108) = 1.394, p = 0.25). Results of this study also demonstrated a completion-bias. Students randomly assigned to Condition B were less likely to complete the problem set than those assigned to Conditions A and C. To conclude, we discuss the counterintuitive findings observed in this study and implications related to developing and implementing feedback in online learning environments for middle school mathematics.