In an observational learning environment, rational agents with incomplete information may mimic the actions of their predecessors even when their own signal suggests the opposite. This herding behavior may lead society to an inefficient outcome if the signals of the early movers happen to be incorrect. This paper analyzes the effect of signal accuracy on the probability of an inefficient informational cascade. The literature so far has suggested that an increase in signal accuracy leads to a decline in the probability of inefficient herding, because the first movers are more likely to make the correct choice. Indeed, the results in Bikhchandani, Hirshleifer and Welch (1992) support this proposition. Here we show that this is not the case in general. We present simulations which demonstrate that even a small departure from symmetry in signal accuracy may lead to non-monotonic results. An increase in signal accuracy may result in a higher likelihood of an inefficient cascade.