A popular approach to interactive effects panel data models is the common correlated effects (CCE) estimator of Pesaran (Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica 74, 967-1012, 2006). The current paper proposes a modified version of this estimator that is useful in a number of cases where the original is not expected to work, such as when the number of cross-sectional units is small. The idea is to use time instead of cross-sectional averages of the observables to purge the interactive effects.