Using a unique dataset linking preschool blood lead levels (BLLs), birth, school, and detention data for 120,000 children born 1990-2004 in Rhode Island, we estimate the impact of lead on behavior: school suspensions and juvenile detention. We develop two instrumental variables approaches to deal with potential confounding from omitted variables and measurement error in lead. The first leverages the fact that we have multiple noisy measures for each child. The second exploits very local, within neighborhood, variation in lead exposure that derives from road proximity and the de-leading of gasoline. Both methods indicate that OLS considerably understates the negative effects of lead, suggesting that measurement error is more important than bias from omitted variables. A one-unit increase in lead increased the probability of suspension from school by 6.4-9.3 percent and the probability of detention by 27-74 percent, though the latter applies only to boys.