Professionals routinely make consequential decisions for others without systematically observing how those decisions turn out. Whether such decisions reflect accurate forecasting or systematically distorted beliefs has important implications for equality of opportunity. We study this question in the context of teachers’ recommendations for high school tracks by combining nationwide administrative data, randomized vignette experiments, belief elicitation, and a field experiment. Holding students’ academic performance and interests constant, teachers are significantly less likely to recommend demanding academic tracks when students come from disadvantaged socioeconomic backgrounds. This differential treatment is accompanied by systematically inaccurate beliefs: teachers substantially overestimate the probability that high-achieving disadvantaged students will fail in academically demanding schools. We then provide teachers with personalized feedback on their previous recommendations and objective information on the realized success of disadvantaged students. The intervention substantially reduces socioeconomic gaps in recommendations for high-achieving students. Effects are concentrated among boys assigned to teachers with the largest prior recommendation gaps and the most inaccurate beliefs. Among the high-achieving disadvantaged boys most affected by the intervention, these changes translate into higher enrollment in demanding academic tracks, without detectable adverse effects on short-run academic performance.