Procedures for handling attrition and missing data in longitudinal studies are discussed. A multiple imputation (MI) strategy is developed that can be applied to complex multilevel data. It is both general and statistically efficient and estimation software is available. An example of its use is given.