The present paper suggests a new way to carry out IV estimation with many instruments. Our suggestion is to cross-sectionally average the instruments and use these averages as instruments. We provide a theoretical and Monte Carlo analysis of this approach.