Academia, and economics in particular, faces increased scrutiny because of gender imbalance. This paper studies the job market for entry-level faculty positions. We employ machine learning methods to analyze gendered patterns in the text of 9,000 reference letters written in support of 2,800 candidates. Using both supervised and unsupervised techniques, we document widespread differences in the attributes emphasized. Women are systematically more likely to be described using "grindstone" terms and at times less likely to be praised for their ability. Given the time and effort letter writers devote to supporting their students, this gender stereotyping is likely due to unconscious biases.