(*Joint with [Jason Tayawa](https://sites.google.com/view/jasontayawa/)*) We experimentally analyze how overconfidence in one's relative expertise impacts social learning in the lab. We modify the social learning settings of Anderson and Holt (1997) to allow for signal accuracy to vary with an individual's skill. Our results show that overconfidence can affect welfare in social learning settings by inducing people to break away from existing herds.