Purpose: Many studies have explored school administrator discretion in the implementation of teacher evaluation and observation systems. However, we are unaware of any studies that quantitatively link discretionary administrator behaviors to student outcomes. The purpose of this study was to (a) explore the determinants of observations arising from administrator discretion and (b) explore the extent to which "discretionary observations" were associated with average student achievement scores and disciplinary offenses. Method: We applied multilevel modeling to 3 years of teacher panel data from more than 80% of Tennessee school districts. Findings: Observable characteristics, differences between schools each year, and teacher traits explain more than 80% of the variation in discretionary observations; teacher prior-year observation and composite effectiveness scores were the strongest predictors. No evidence suggested that average student achievement scores or behavior suffered among teachers who received fewer observations than assigned by policy. Average student achievement scores rose among teachers receiving supplementary observations compared with the years when they received the policy-prescribed number. Implications for Research and Practice: Quantitative research can substantially control for discretionary observations using multilevel modeling. Observers might be encouraged to not observe high-performing teachers more than what is prescribed by policy. Observers seemingly deviate from observation policy in ways that do not increase office referrals and may improve student achievement.