Background: Developments in educational technology and learning analytics make it possible to automatically formulate and deploy personalized formative feedback to learners at scale. However, to be effective, the motivational and emotional impacts of such automated and personalized feedback need to be considered. The literature on feedback suggests that effective feedback, among other features, provides learners with a standard to compare their performance with, often called a reference frame. Past research has highlighted the emotional and motivational benefits of criterion-referenced feedback (i.e., performance relative to a learning objective or mastery goal) compared to norm-referenced feedback (performance relative to peers). Objectives: Despite a substantial body of evidence regarding reference frame effects, important open questions remain. The questions encompass, for example, whether the benefits and drawbacks of norm-referenced feedback apply in the same way to automated and personalize feedback messages and whether these effects apply to students uniformly. Further, the potential impacts of combining reference frames are largely unknown, even though combinations may be quite frequent in feedback practice. Finally, little research has been done on the effects of reference frames in computer-supported collaborative learning, which differs from individual learning in meaningful ways. This study aims to contribute to addressing these open questions, thus providing insights into effective feedback design. Specifically, we aim to investigate usefulness perceptions as well as emotional and motivational effects of different reference frames--and their combination--in automated and personalized formative feedback on a computer-supported collaborative learning task. Methods: A randomized field experiment with four feedback conditions (simple feedback, norm-referenced, criterion-referenced, and combined feedback) was conducted in a course within a teacher training program (N = 282). Collaborative groups worked on a learning task in the online learning environment, after which they received one of four possible automated and personalized formative feedback. We collected student data about feedback usefulness perceptions, motivational regulation, and achievement emotions to assess the differential effects of these feedback conditions. Results: All feedback types were perceived as useful relative to the simple feedback condition. Norm-referenced feedback showed detrimental effects for motivational regulation, whereas combined feedback led to more desirable motivational states. Further, criterion-referenced feedback led to more positive emotions for overperformers and to more negative emotions for underperformers. The findings are discussed in light of the broader feedback literature, and recommendations for designing automated and personalized formative feedback messages for computer-supported collaborative learning are presented.