In this study, we employ an innovative new methodology suggested by Bernhardt et al. (2006) to examine the herding (or anti-herding) behavior of German analysts in case of earnings forecasts. Our methodology avoids well-known shortcomings often encountered in related studies, such as, e.g., correlated information signals, unexpected common shocks to earnings, systematic optimism or pessimism or forecast target mismeasurement. Our findings suggest that German analysts anti-herd, i.e., they systematically issue earnings forecasts which are further away from the consensus forecast than their private information indicates. Furthermore, we analyze the association between herding behavior and different characteristics which might influence analysts' herding behavior, like the size of the brokerage, the experience of the analyst and the coverage of firms of the Neuer Markt. We mainly confirm findings for the U.S., e.g., that the anti-herding phenomenon is more severe in cases of higher competition among analysts. Contrary to anecdotal evidence, we also find anti-herding behavior when analysts issue forecasts for Neuer Markt firms during the 'new economy' bubble.