This paper presents an empirical study of how ranging affects multihop localization in wireless sensor networks. We compare real-world deployments with simulation and establish an objective metric with which to evaluate ranging models: if a model accurately predicts real-world localization performance, it accurately captures ranging characteristics. We evaluate a well-established ranging model called Noisy Disk and find that it yields optimistic predictions; real deployment error can be four to five times worse than simulation error. A systematic analysis reveals that Noisy Disk fails to capture important connectivity characteristics which dominate the effects of error on localization.