This paper investigates the effects of neighborhood size and network structure on strategic experimentation. We analyze a multi-arm bandit game with one safe and two risky alternatives. In this setting, risk taking produces a learning externality and an opportunity for free riding. We conduct a laboratory experiment to investigate whether group size and the network structure affect risk taking. We find that group size has an effect on risk taking that is qualitatively in line with equilibrium predictions. Introducing an asymmetry among agents in the same network with respect to neighborhood size leads to substantial deviations from equilibrium play. Findings suggests that subjects react to changes in their direct neighborhood but fail to play a best-response to their position within the network.