While herding has long been suspected to play a role in financial market booms and busts, theoretical analyses have struggled to identify conclusive causes for the effect. Recent theoretical work shows that informational herding is possible in a market with efficient asset prices if information is bi-polar, and contrarianism is possible with single-polar information. We present an experimental test for the validity of this theory, contrasting with all existing experiments where rational herding was theoretically impossible and subsequently not observed. Overall we observe that subjects generally behave according to theoretical predictions, yet the fit is lower for types who have the theoretical potential to herd. While herding is often not observed when predicted by theory, herding (sometimes irrational) does occur. Irrational contrarianism in particular leads observed prices to substantially differ from the efficient benchmark. Alternative models of behavior, such as risk aversion, loss aversion or error correction, either perform quite poorly or add little to our understanding.