Explains why methods such as ecological regression so often fail and examines the promising new techniques for cross-level inference. The authors argue for a range of alternate techniques, including extensions of cross-tabular, regression analysis and unobservable variable estimators.